Social Network Analysis of Information Flow and Opinion Formation on Indonesian Social Media: A Case Study of Youth Violence

Abstract

This study examines the dynamics of information dissemination and opinion formation in Indonesian social media through a comprehensive analysis of a high-profile youth violence case. Using social network analysis (SNA), we analyzed 264,155 activities from 83,097 accounts on platform X (formerly Twitter) to understand the patterns of information flow, cluster formation, and inter-group interactions. The analysis revealed four distinct clusters with unique characteristics: a dominant support cluster (40.12%), a context-focused cluster (26.93%), a mainstream media cluster (14.14%), and a peripheral engagement cluster (6.05%). This study found significant patterns in information dissemination, with retweets dominating at 68% of total activities and strategic hashtag usage at 28%. Cross-cluster interactions comprised 20% of total activities, challenging assumptions about echo chambers in digital discourse. The network showed high resilience with 85% path reliability and demonstrated a consistent multiplier effect with a 1:5:15 ratio in message amplification. Bridge nodes (10–15% of accounts) played crucial roles in facilitating cross-cluster dialogue and maintaining network cohesion. The temporal evolution of discourse showed distinct phases, from initial factual reporting to later systemic analysis, with each phase characterized by different engagement patterns and narrative focuses. These findings extend existing theoretical frameworks while highlighting the need for more culturally nuanced approaches to understanding digital discourse in contexts of collectivist cultural dimensions. This study’s results have significant implications for digital literacy education, social media intervention strategies, and youth violence prevention efforts, suggesting the need for sophisticated, network-aware approaches that consider both structural dynamics and cultural contexts.

1. Introduction

The digital transformation of public discourse has fundamentally changed how society interacts and communicates about sensitive social issues. In Indonesia, with 139 million active social media users out of a 278 million population [1], digital platforms have become the primary arenas for public opinion formation. However, this increased digital access has also led to various information disorders, including the spread of misinformation and hate speech in cases involving youth violence.

Throughout 2023–2024, several youth violence cases captured public attention on Indonesian social media, with the Mario Dandy case emerging as a particularly significant example of how digital platforms shape public discourse around youth violence. This case, involving physical violence by a prominent official’s son that was recorded and widely shared, triggered extensive public debate about privilege and parenting, generating substantial social media engagement and revealing complex patterns of information flow and opinion formation.

Social media’s impact on youth violence has become increasingly concerning in Indonesia. UNICEF surveys found that one in three adolescents has experienced cyberbullying [2]. Recent data show an alarming 23% increase in youth violence cases over the past two years, with 60% of these cases directly linked to social media activities, either as triggers, documentation media, or information dissemination channels [3].

The phenomenon of “Fight Compilations” on platforms like YouTube has exacerbated the situation, with real fights being compiled and widely distributed. Studies show that repeated exposure to violent content correlates with antisocial behavior and desensitization to aggression [4]. Adolescents who regularly consume such content tend to internalize violence as a conflict resolution method, reflecting Social Cognitive Theory principles [5,6].

A systematic study analyzing 56 articles (2001–2013) on the relationship between youth violence and online behavior revealed that exposure to online threats, such as cyberbullying and gang activities, can predict offline risk behavior [7]. In Indonesia, research shows similar correlations between online activities and increased tendencies toward aggressive behavior [3].

The emotional resonance of misinformation plays a crucial role in shaping perceptions of youth violence. Content exploiting fear and anxiety drives acceptance and the spread of violent rhetoric [8]. Research in Indonesia has identified how hate narratives and disinformation on social media contribute to youth conflict escalation [9,10].

The impact of online violence exposure extends beyond individuals, encompassing disrupted decision-making and increased mental health risks [5,11]. Studies in Indonesia reveal how cyberbullying not only affects victims’ mental health but can also trigger more serious physical violence [2].

This study focuses on four key research questions:

• Who are the key actors involved in narrative formation and dissemination?
• What patterns characterize information dissemination in youth violence cases on Indonesian social media?
• How do opinion clusters form and evolve within digital social networks?
• What are the characteristics of inter-group interactions within the network?

While the Mario Dandy case represents a specific instance of youth violence involving privileged individuals, it serves as a valuable representative case for several reasons. First, it exemplifies a pattern of similar cases in Indonesia where children of public officials engage in violent behavior, often feeling a sense of impunity due to their parents’ positions. During our research period alone, comparable cases emerged, including that of the son of police official AKBP Achiruddin Hasibuan, demonstrating that this is not an isolated phenomenon but part of a recurring pattern.

Second, the case illustrates the increasing intersection between youth violence and digital media that characterizes contemporary Indonesian society. UNICEF surveys indicate that one in three Indonesian adolescents experiences cyberbullying [2], while recent data show a 23% increase in youth violence cases over the past two years, with 60% directly linked to social media activities [3]. The Mario Dandy case, with its digital documentation and viral spread, exemplifies this growing connection between physical violence and digital spaces.

Third, the case’s extraordinary virality on social media makes it particularly valuable for understanding how digital networks shape public discourse around youth violence. By examining a case that generated exceptionally high engagement, we can observe discourse patterns more clearly than would be possible with cases receiving minimal digital attention. The case serves as a “critical incident” that brings typically submerged social dynamics into sharp relief, making it methodologically valuable despite its unique characteristics.

Finally, while acknowledging the case’s distinctive aspects involving elite privilege, we position it within broader patterns of youth violence in Indonesia while maintaining appropriate caution regarding generalizability. This approach allows us to extract valuable insights about digital discourse formation while recognizing the specific socio-political context that influenced the case’s trajectory.

The significance of this research lies in its methodological, practical, and theoretical contributions. Methodologically, it employs social network analysis (SNA) to reveal hidden patterns in information dissemination and public opinion formation. Practically, the findings can inform the development of more effective digital literacy programs and policy interventions for addressing youth violence in the digital era. Theoretically, it contributes to understanding the transformation of Indonesia’s digital public sphere, exploring how information disorders shape collective understanding and social responses to youth violence issues.

This study focuses specifically on the platform X (formerly Twitter) due to its unique role in the Mario Dandy case. While Instagram and TikTok have larger user bases among Indonesian youth according to the We Are Social report [1], Twitter emerged as the critical platform where this case gained extraordinary traction and ultimately influenced institutional responses. The case exemplifies digital activism’s power through Twitter, as the platform’s text-centric nature and conversation-oriented features facilitated substantive public discourse that transcended mere viral sharing.

Our preliminary analysis revealed that public pressure generated through Twitter directly influenced mainstream media coverage and institutional responses, with official investigations accelerating following intense social media scrutiny. The Tax Directorate and Ministry of Finance’s decisions to investigate Rafael Alun Trisambodo came after his son’s case went viral on Twitter, demonstrating the platform’s unique capacity to transform individual incidents into issues of institutional accountability. This distinctive role of Twitter in shaping both public discourse and institutional response makes it the most appropriate platform for examining how digital networks influence public opinion formation in cases of youth violence with broader social implications.

In February 2023, Mario Dandy Satriyo (MDS), the son of Rafael Alun Trisambodo or RAT (an employee of the Directorate General of Taxes, Indonesian Ministry of Finance), brutally assaulted David Ozora (DO) in Pesanggrahan, South Jakarta. The motive for the persecution stemmed from a personal conflict in which David Ozora was allegedly still trying to get close to his ex-girlfriend, AGH (initials used because she is still a minor), who at that time had become MDS’s lover. Fueled by jealousy and anger, Mario and his friend Shane Lukas (SL) arranged a meeting with David, which then turned into an act of violence.

The abuse committed by MDS against DO was recorded by SL using a cell phone. DO suffered severe injuries including brain damage and had to be intensively hospitalized. The video footage was then leaked and began circulating on instant messaging apps before eventually spreading to Twitter (X), going viral, and sparking public outrage.

As the case went viral on social media, netizens began digging up information about MDS and her family. It was revealed that the family’s lavish lifestyle did not align with the income of a tax official, including a collection of luxury cars, as well as overseas trips and expensive branded goods. Hashtags related to this case quickly became a trending topic on Twitter, prompting the police to act swiftly to arrest MDS and SL.

Growing public pressure eventually compelled the Ministry of Finance and the Directorate General of Taxes to carry out an internal audit of the RAT’s wealth. Consequently, MDS’s father was dismissed from his position as a tax employee and underwent a tax audit concerning alleged tax evasion. Meanwhile, MDS and SL faced charges related to serious and premeditated maltreatment and child protection.

This case captured widespread public attention not only due to the brutality of the violence but also because it intersected with broader social issues in Indonesian society: wealth inequality, corruption among public officials, and the perception of impunity for the wealthy. The case revealed how social media served both as a vector for documenting violence and as a powerful tool for public accountability. As the case progressed through the legal system from February to December 2023, it continued to generate significant engagement on social media, with diverse narratives emerging around issues of justice, privilege, and social class. These characteristics make the Mario Dandy case particularly valuable for analyzing how digital discourse surrounding youth violence evolves, especially when intersecting with issues of power and social inequality.

1.1. Digital Public Sphere and Network Society Theory

The development of social media has fundamentally transformed how society interacts and communicates, creating new forms of public discourse and social organization. Castells [12], in his seminal work “Networks of Outrage and Hope”, provides a comprehensive theoretical framework for understanding this transformation. He argues that contemporary society has shifted from traditional hierarchical structures toward more complex and fluid network architectures, where power flows through communication networks that enable real-time social coordination and mobilization rather than being centered in formal institutions.

Building on this foundation, Boyd [13] deepens our understanding by asserting that social media must be understood as a socio-technical phenomenon, not merely a collection of communication technologies. She explains how social media emerged as a response to the dot-com bubble collapse, creating a new paradigm in how technology mediates social interaction. This perspective enriches our understanding of how digital platform practices become embedded in contemporary social life, transcending their technical function as communication tools.

This transformation of the digital public sphere is further elaborated on by van Dijck et al. [14] through their critical analysis of the “platform society”. They demonstrate how digital platforms not only facilitate communication but also actively shape and regulate social interaction through algorithmic mechanisms. Their analytical framework reveals how the platform architecture and algorithmic governance create new forms of social organization and control, fundamentally altering how information flows and public opinion forms.

In this context, Stray et al. [15] highlight how polarization, violence, and social media are interconnected. They critique social media platforms’ approach, which relies heavily on content moderation, arguing that moderation only affects a small portion of content that objectively violates policies, and that expanding moderation efforts can lead to more bias in enforcement and controversy.

1.2. Youth Violence and Digital Networks

The intersection of social media and youth violence presents particular challenges in the digital age. Patton et al. [11] identify how social media can become vectors for various forms of violence, including cyberbullying, gang violence, and aggression in romantic relationships. Their research reveals that while the majority of adolescents (65–91%) report minimal involvement in social media violence, digital platforms have created new spaces for conflict escalation and amplification.

This vulnerability is further complicated by what Livingstone and Third [16] describe as the unique position of adolescents in the digital landscape—they are simultaneously pioneers and vulnerable subjects. As digital natives, teenagers often lead in adopting new technologies, yet their developmental stage makes them particularly susceptible to online risks. Sultan et al. [17] reinforce this concern, noting that teenagers’ significant time investment in social media raises major concerns among parents and policymakers about its impact on healthy development.

The complexity of these challenges is illuminated by Chan et al. [18], who presented an examination of cyberbullying experienced by teenagers. Cyberbullying itself represents an umbrella term encompassing various forms of online aggression with distinct characteristics and impacts. Scholars have categorized these manifestations into several typologies, each with unique psychological and social implications [19,20]. Direct cyberbullying includes harassment (the persistent sending of offensive messages), flaming (heated, hostile exchanges), and denigration (spreading rumors to damage a reputation), all characterized by the explicit targeting of victims. Indirect forms encompass exclusion (deliberately ostracizing someone from online groups), impersonation (assuming a victim’s digital identity to damage relationships), outing (sharing private information without consent), and cyberstalking (persistent monitoring and unwanted contact causing fear) [18,21].

These various forms of cyberbullying have different prevalence rates and psychological impacts. A meta-analysis by Kowalski et al. [19] found that denigration and harassment were the most common forms (occurring in 60–75% of cyberbullying incidents), while more severe forms, like impersonation and cyberstalking, though less frequent (15–25% of incidents), often produced more serious psychological distress. These challenges are further exacerbated by the distinctive characteristics of digital media—anonymity enabling disinhibition, constant accessibility eliminating safe spaces, and the potential for rapid content virality, which can transform isolated incidents into widespread public humiliation [22,23]. In the Indonesian context specifically, studies by Bukhori et al. [24] and Safaria [25] have documented how these various forms manifest with particular cultural dimensions, including the use of religion-based harassment and family honor threats as distinctly localized cyberbullying tactics. These cultural elements add complexity to addressing cyberbullying in Indonesia’s collectivist society.

Adding another layer to this complexity, Tomassi et al. [26] and Wardle and Derakhshan [27] analyze the phenomenon of information disorder in social media. They identify three main forms: misinformation (unintentionally false information), disinformation (deliberately manipulated information), and malinformation (distorted true information). Building on platform society concepts, Stray et al. [15] suggest the need for more proactive and long-term approaches to managing digital conflict, operating through platform design rather than solely relying on content moderation policies. This framework helps reveal how different types of problematic information can affect youth behavior and perception in digital spaces.

Recent developments in understanding misinformation dynamics provide further theoretical context for analyzing youth violence cases on social media. Wardle [28] argues that the traditional focus on categorizing content as misinformation or disinformation is inadequate for grasping the complex dynamics of online discourse. She suggests moving beyond the “atoms of content” approach to examine broader narratives and user motivations. According to Wardle [28], individuals are influenced less by individual posts and more by the collective narratives these posts create, much like how repeated drops of water can eventually carve grooves in stone. This perspective enhances the Information Disorder Framework [27] by highlighting the social contexts in which information spreads and the identity-based motivations driving content sharing.

1.3. Network Analysis in Indonesian Digital Context

Understanding these digital phenomena requires sophisticated analytical approaches, particularly in specific cultural contexts like Indonesia. Monge and Contractor [29] offer a comprehensive framework for analyzing the complexity of digital communication networks, which has proven valuable in such contexts. Their multitheoretical approach enables a more nuanced understanding of how various levels of analysis—from micro-interactions to macro-structures—interact in shaping communication dynamics. This framework is particularly relevant when examining how information flows and opinions form in culturally specific digital environments.

Building on this analytical foundation, Borgatti and Halgin [30] develop models explaining how information and influence flow through digital social networks. Their work helps understand why some narratives or ideas can spread rapidly while others remain limited, a particularly important consideration in Indonesia’s diverse and dynamic social media landscape. The models they propose help explain how cultural and social factors influence the velocity and reach of information in digital networks.

The application of these network analysis approaches in Indonesia reveals unique patterns shaped by local cultural dynamics. Lim [31] demonstrates that communication patterns on Indonesian social media are heavily influenced by cultural characteristics and established social networks. Her research shows that information dissemination in Indonesian digital spaces often depends on pre-existing social networks and communities, reflecting how traditional communication patterns adapt to and manifest in digital environments. This finding underscores the importance of considering the cultural context when analyzing digital networks.

The broader Southeast Asian context, as analyzed by Sinpeng [32], provides additional insights into how social media functions within specific cultural and political frameworks. While digital technology enables more effective mobilization and coordination, various social and institutional factors influence how these capabilities are realized. This regional perspective helps situate Indonesia’s digital transformation within a broader cultural and political context.

Recent research by Anom et al. [33] further enriches our understanding of Indonesia’s unique digital dynamics through their study of Generation Z’s information consumption patterns. Their analysis during the COVID-19 pandemic reveals that despite being digital natives, young Indonesians’ patterns of consumption and information validation remain strongly influenced by cultural values. This finding highlights the persistent role of traditional social values in shaping digital behavior, even among the most technologically adept generations.

These theoretical and analytical perspectives converge to provide a robust framework for understanding how digital networks function in culturally specific contexts. The integration of network analysis approaches with understanding of local cultural dynamics enables a more accurate and nuanced analysis of how information flows and opinions form in Indonesian digital spaces. This is particularly relevant when studying cases of youth violence and social conflict that manifest in and through social media platforms.

Collectively, this literature review demonstrates the need for approaches that combine sophisticated network analysis with deep understanding of the cultural context when studying digital phenomena in Indonesia. This integrated perspective informs our methodology and analysis of how youth violence cases are discussed and understood in Indonesian social media networks.

Recent scholarship has expanded our understanding of platform governance beyond technical content moderation approaches. Gowder [34], in “The Networked Leviathan”, introduces a democratic framework for understanding digital platforms, arguing that they suffer from a fundamental “democratic deficit” due to their lack of accountability to the publics they serve. Unlike Gosztonyi’s [35] focus on content moderation mechanics, Gowder emphasizes the necessity of “governance entities at multiple scales” and greater empowerment for platform users and workers.

Gowder’s framework is especially relevant for analyzing cross-cultural conflicts on social media, as he highlights how platform rules often inadequately address cultural contexts that contribute to digital violence. He points out that large platforms like Meta face significant challenges in consistently enforcing their own rules, resulting in inconsistencies in managing hate speech and online violence. This perspective provides valuable insights for examining how cases of youth violence on Indonesian social media are moderated within globally operated platforms that may lack sufficient cultural context for effective governance.

Furthermore, Gowder [34] explores the role of nation-states in platform regulation, suggesting that while democratic platform governance should not solely depend on nation-states, governments still play important roles in providing appropriate interventions through human rights policies and competition regulations. This multi-level governance perspective complements Gosztonyi’s analysis of content moderation systems and offers a broader framework for understanding how digital discourse about youth violence is shaped by governance structures at various levels.

1.4. Theoretical Framework

This study adopts an integrated theoretical framework that combines several key perspectives to analyze digital discourse dynamics around youth violence on Indonesian social media. Each theoretical component provides distinct but complementary tools for understanding different aspects of digital communication networks and their social implications.

The foundation of our theoretical framework rests on Network Society Theory, developed by Castells [12], which fundamentally reconceptualizes how power operates in contemporary digital society. Castells argues that society has shifted from traditional hierarchical structures to network architectures, where power flows through communication networks rather than residing in fixed institutional positions. This transformation has created what Castells terms “communication power”—the ability to influence social processes through control of communication networks, which is particularly relevant when analyzing how narratives about youth violence form and spread in digital spaces.

Building on this foundation, the Platform Society concept elaborated by van Dijck et al. [14] provides crucial insights into how digital platforms actively shape these communication networks. Their framework identifies three key mechanisms through which platforms influence communication:

• Datafication: The transformation of social activity into quantified data.
• Commodification: The conversion of social values into economic values.
• Selection: The algorithmic curation of information and interaction.

These mechanisms help explain how platform architecture and algorithmic governance create specific conditions under which discussions about youth violence evolve and spread.

While platform mechanisms shape the structure of communication, the content and quality of information flowing through these networks demand additional theoretical tools for analysis. In this regard, the Information Disorder Framework developed by Wardle and Derakhshan [27] and expanded by Tomassi et al. [26], as discussed in Section 1.2, becomes particularly valuable. This framework is crucial for understanding how various forms of problematic information impact public discourse surrounding youth violence cases, especially in the emotionally charged context of social media discussions.

To systematically analyze these complex network dynamics, we employ Monge and Contractor’s [29] Digital Communication Networks analysis framework. Their multitheoretical approach enables an examination at multiple levels—from micro-level individual interactions to macro-level network structures.

• Micro-level: Individual interaction patterns.
• Meso-level: Group formation and cluster dynamics.
• Macro-level: Overall network structures and evolution.

This allows for a systematic analysis of how information flows and opinions form within digital networks. This methodological framework provides the tools needed to understand how information flows and opinions form within digital networks, complementing the theoretical insights from Castells [12] and van Dijck [14].

These global theoretical perspectives are then grounded in the specific context of Indonesian digital culture through the work of local scholars. Lim’s [31] research on Indonesian social media patterns and Anom et al.’s [33] analysis of Generation Z’s digital behavior provide crucial insights into how global digital phenomena manifest in specific cultural settings. This local contextualization is essential for understanding how universal patterns of digital communication adapt to and are shaped by Indonesian social and cultural factors.

The integration of these theoretical perspectives enables a comprehensive analysis of four key dimensions:

• Network Formation: How digital communities form and evolve around youth violence cases.
• Information Flow: How different types of content spread through social networks.
• Opinion Dynamics: How public discourse develops and changes over time.
• Cultural Context: How Indonesian social and cultural factors influence digital interaction patterns.

By combining these theoretical perspectives, we can better understand the complex interplay between digital platform dynamics, information flows, and social responses to youth violence on Indonesian social media.

2. Materials and Methods

2.1. Methodological Framework

This study employs a quantitative approach using a social network analysis (SNA) methodology to analyze interaction patterns and information dissemination on social media. The selection of SNA as the primary methodology is based on its capability to reveal relationship structures and communication patterns in digital social networks [36,37]. In the context of digital communication research, SNA enables researchers to understand not only communication content but also the structure and dynamics of interactions occurring between social media users.

2.2. Data Collection and Processing

Data were collected from the platform X (formerly Twitter) using the platform’s official API, ensuring data authenticity and reliability (see data availability at 10.5281/zenodo.14890595). For the Mario Dandy case (1 February–31 December 2023), a simple query “mario dandy” was chosen due to the case’s specific and unique name. The extended time range was selected to cover case developments from the initial incident through legal proceedings.

The selection of the platform X (formerly Twitter) as our primary data source requires specific justification, particularly as other platforms like Instagram and TikTok have greater overall reach among Indonesian youth. Despite not being the most widely used social media platform in Indonesia, X was deliberately chosen for several methodological and substantive reasons directly relevant to the Mario Dandy case.

First, X served as the primary arena where this case evolved from individual violence to a national conversation about privilege and institutional accountability. Our preliminary data showed extraordinary engagement metrics for this case on X, with individual tweets receiving up to 335K engagements—levels rarely seen for youth violence cases. The platform’s architecture enabled public pressure, which demonstrably influenced institutional responses, with official investigations by the Tax Directorate appearing to accelerate following intense Twitter scrutiny.

Second, X’s structural characteristics support more sophisticated discourse analysis than visually oriented platforms. The platform’s follower–following structure, combined with its conversation threading, hashtag functionality, and public reply chains, creates traceable interaction patterns that allow for robust network analysis. These features enabled us to identify distinct discourse communities and interaction patterns that might be less visible on platforms with different architectural designs.

Third, X’s API offered more comprehensive access to historical data during our research period, allowing us to gather a complete dataset across the entire case timeline (February–December 2023). This enabled an analysis of the temporal evolution in discussion patterns, which would have been methodologically challenging on platforms with more restricted data access.

Finally, while Instagram and TikTok have larger user bases among younger Indonesians, X represents a critical platform for opinion formation among journalists, policy influencers, and public commentators who shape broader media narratives. In the Mario Dandy case specifically, discourse that originated on Twitter subsequently influenced coverage in mainstream media outlets and ultimately triggered institutional responses, demonstrating its unique position at the intersection of social media activism, journalism, and public policy.

The data-cleaning process involved several systematic steps:

• Tweet Duplication Removal
  • Identifying and removing identical tweets appearing multiple times;
  • Checking based on tweet ID and content;
    • This step was crucial because the platform X’s API occasionally returns duplicate content, especially during high-traffic periods. We identified duplicates through tweet ID comparison and content similarity analysis, utilizing Python’s library to filter unique entries while maintaining the integrity of the conversation threads.
• Format Standardization
  • Converting date and time to same time zone;
  • Consistent formatting of usernames and mentions;
  • Hashtag normalization to address capitalization variations.
    • This involved converting all timestamps to Western Indonesia Time (WIB) to accurately track temporal patterns, as Indonesia spans multiple time zones and many users’ devices might be set to different configurations. We also standardized the formatting of usernames and mentions by removing special characters and normalizing capitalization variations to avoid counting the same user multiple times under slightly different renderings of their username. Hashtag normalization was equally important, as variations in capitalization (e.g., #JusticeForDavid versus #justicefordavid) could fragment our analysis of thematic patterns. Using regex pattern matching, we standardized all hashtags to lowercase format while preserving their original semantic meaning.
• Data Completeness Verification
  • Essential information checking: user ID, timestamp, tweet content;
  • Engagement metrics;
  • Interaction metadata.
    • Each tweet record was examined for critical data points, including user ID, timestamp, and tweet content. We also verified engagement metrics (retweets, likes, replies) and interaction metadata (hashtags, mentions, media attachments) to ensure a complete picture of each interaction. Records with substantial missing data were flagged and, where possible, reconciled with secondary data sources. This verification process was essential for maintaining data integrity, particularly for the longitudinal analysis of conversation evolution over the case timeline.

2.3. Data Filtering and Categorization

The filtering process included the following:

• Topic Relevance Filtering
  • Manual and automated checking of tweet relevance;
  • Context verification of keyword usage;
  • Removal of off-topic tweets.
    • Topic relevance filtering began with automated keyword-based identification, followed by manual verification by three trained coders. We implemented context verification of keyword usage to distinguish between relevant discussions of the Mario Dandy case and incidental mentions or homonyms. For example, tweets containing “Mario” but referring to the video game character were excluded. This manual verification process, while time-intensive, was crucial for maintaining the integrity of the dataset, as purely automated approaches would have included substantial off-topic content due to the complex nature of the case.
• Spam and Bot Removal
  • Identification based on posting patterns;
  • Account characteristics analysis;
  • Promotional content removal.
    • Spam and bot removal represented a critical quality control measure, particularly due to the case’s high public visibility, which attracted automated engagement. We developed a multidimensional identification approach based on posting patterns, including an unnaturally high posting frequency (exceeding 50 tweets per hour), identical content posted across multiple threads, and repetitive engagement patterns. The analysis of account characteristics included examining the account age, follower–following ratios, and behavioral patterns typical of automated accounts. Additionally, we filtered promotional content that opportunistically used case-related hashtags to market unrelated products or services. This comprehensive approach removed approximately 5783 tweets (2.14% of the initial dataset) identified as likely non-human or spam interactions.
• Data Categorization
  • By interaction type (original tweets, retweets, replies, quotes);
  • By content type (pure text, media, URL, combinations);
  • By user characteristics (verified status, follower count, activity level);
  • By temporal phases.
    • Following these quality control measures, we implemented a systematic data categorization framework to facilitate multidimensional analysis. Tweets were categorized by interaction type to distinguish between original content creation, information amplification (retweets), direct engagement (replies), and evaluative responses (quotes). This categorization was essential for understanding the various communicative functions within the discourse. Content type categorization separated purely textual contributions from those incorporating media (images, videos), URLs (linking to external content), or combinations thereof, enabling the analysis of how different content formats influenced engagement patterns.
    • User characteristic categorization allowed us to examine how the verified status, follower count, and activity level correlated with influence and reach within the network. We created distinct categories for high-influence users (>10,000 followers), medium-influence users (1000–10,000 followers), and standard users (<1000 followers) to analyze how account reach affected information dissemination. Finally, temporal phase categorization organized the dataset into initial (0–24 h post-incident), middle (24–72 h), and advanced (>72 h) phases, enabling the examination of how discourse evolved from immediate reactions to more reflective analysis over time. This multidimensional categorization framework established the analytical foundation for identifying the distinct clusters and interaction patterns presented in our results.

2.4. Network Analysis Approach

The network analysis included the following:

• Structural Analysis
  • Degree centrality calculation for central actor identification;
  • Betweenness centrality analysis for information brokers;
  • Eigenvector centrality measurement for influence;
  • Network density analysis for cohesiveness.
• Cluster and Community Analysis
  • Community identification using Louvain algorithm;
  • Modularity analysis;
  • Clustering coefficient calculation;
  • Inter-community interaction analysis.
• Temporal Analysis
  • Tracking network structural changes;
  • Cluster formation and dissolution analysis;
  • Critical momentum identification;
  • Discussion sustainability analysis.

2.5. Visualization

Network visualization was conducted using Gephi 0.9.2, considering the following:

• Layout and Visual Representation
  • ForceAtlas2 for optimal layout;
  • Visual parameter adjustments;
  • Filter implementation;
  • Label optimization.
• Dynamic Visualization
  • Temporal evolution representation;
  • Structural change animation;
  • Interactive timeline;
  • Temporal filtering.

2.6. Validation and Reliability

Validation protocol, several systematic validation steps were implemented:

• Structural Data Validation
  • Internal consistency checks;
  • Temporal validation;
  • Network structural coherence verification.
• Cross-Source Validation
  • Verification with mainstream media coverage;
  • Data triangulation;
  • Expert consultation.
• Methodological Validation
  • Inter-coder reliability testing;
  • Statistical validation;
  • Robustness checks.

3. Results

3.1. Network Characteristics and Volume

Analysis of the Mario Dandy case (refer to Figure 1) revealed extensive public engagement, with 264,155 activities generated by 83,097 accounts during the study period. This significant volume of engagement demonstrates the case’s substantial impact on Indonesian social media discourse, with each participating account generating an average of 3.18 activities, indicating sustained rather than merely casual involvement in the discussion.

The distribution of activities showed distinctive patterns that reveal how information spread and how users engaged with the topic. Retweets overwhelmingly dominated the interaction landscape, with 182,308 instances, comprising 68% of total activities. This striking predominance of retweets indicates a clear “share first, discuss later” behavior pattern among users, where information amplification takes precedence over original content creation or in-depth discussion. Such behavior aligns with typical viral discourse characteristics on social media, where the urgency to spread information outweighs the inclination to engage in substantive dialogue.

Hashtag usage emerged as the second most common activity, with 74,947 occurrences representing 28% of total activities. More significantly, the presence of 3534 unique tags demonstrates systematic efforts by users to organize and categorize discussions. This sophisticated use of hashtags suggests a deliberate attempt to not only increase content visibility but also to create structured conversations around specific aspects of the case. The variety and volume of hashtag usage indicate coordinated efforts to maintain the topic’s prominence while simultaneously categorizing different narrative threads within the broader discussion.

In contrast, dialogic interactions showed notably lower volumes, with replies (5629|2%), mentions (2543|1%), and quoted tweets (2062|1%) collectively accounting for only 4% of total activities. This significant disparity between content amplification and actual discussion reveals a crucial pattern in how social media users engage with sensitive topics. The minimal proportion of dialogic interactions, despite the high overall engagement, suggests that while users were heavily invested in spreading information about the case, they were markedly less inclined to engage in direct conversations or critical discussions about its implications.

This distribution pattern provides important insights into the nature of social media engagement with sensitive topics like youth violence. The overwhelming preference for retweets combined with sophisticated hashtag usage but limited dialogic interaction suggests a complex engagement pattern where users actively participate in information dissemination and organization but shy away from direct discussion. This behavior might reflect both the sensitive nature of the topic and broader patterns of social media usage, where quick, low-effort forms of engagement are preferred over more demanding forms of interaction that require original thought and potential exposure to controversy.

3.2. Community Structure and Polarization

The analysis identified four distinct clusters with unique characteristics and roles (refer to Figure 2):

• Pro David cluster (33,338 accounts|40.12%)
  • Highest internal cohesion (85% internal activities);
  • Strong narrative consistency supporting the victim;
  • Most active in cross-cluster interactions (15,001 interactions).
• Mario Dandy cluster (22,378 accounts|26.93%)
  • More balanced interaction pattern (75% internal, 25% external);
  • Focus on perpetrator background and social context;
  • Significant interaction with media cluster (5594 interactions).
• Media cluster (11,750 accounts|14.14%)
  • Neutral position in information dissemination;
  • Highest content adoption rate;
  • Balanced interaction distribution with other clusters.
• Buzzer cluster (5027 accounts|6.05%)
  • Highest activity per account (3.54);
  • Opportunistic engagement patterns;
  • Limited substantive contribution (5% of direct interactions).

The analysis revealed a complex network structure organized into four distinct clusters, each exhibiting unique characteristics and playing specific roles in the overall discourse. These clusters emerged naturally through interaction patterns and demonstrated varying levels of engagement, influence, and narrative focus.

The largest and most influential group was the Pro David cluster, comprising 33,338 accounts (40.12% of total users). This cluster demonstrated the highest internal cohesion, with 85% of activities occurring within the cluster itself, indicating strong group solidarity and narrative alignment. Despite this high internal cohesion, the cluster was also the most active in cross-cluster interactions, generating 15,001 interactions with other groups, suggesting a confident and outward-reaching approach to spreading its victim-supporting narrative.

In contrast to this victim-focused group, the Mario Dandy cluster emerged as the second-largest group, with 22,378 accounts (26.93%). This cluster exhibited a more balanced interaction pattern, with 75% internal activities and 25% external engagement, indicating a greater openness to dialogue with other perspectives. Its focus centered on examining the perpetrator’s background and broader social context, maintaining significant interaction with the Media cluster (5594 interactions) to validate and contextualize its discussions.

The Media cluster, while smaller, at 11,750 accounts (14.14%), played a crucial bridging role in the network. This cluster maintained a neutral position in information dissemination and achieved the highest content adoption rate among all groups. Its balanced interaction distribution with other clusters positioned it as an important validator and information hub, facilitating fact-based discussions across different perspectives.

The smallest but notably active group was the Buzzer cluster, consisting of 5027 accounts (6.05%). Despite its limited size, this cluster recorded the highest activity per account at 3.54 interactions, significantly above the network average. However, its engagement pattern was largely opportunistic, characterized by high-volume but low-impact activities. With only 5% of direct interactions contributing to substantive discussion, its role appeared more focused on amplification than meaningful dialogue.

These clusters formed not just as opinion groups but as distinct communities with unique interaction patterns and narrative approaches. The Pro David cluster emerged as the dominant force in both size and cross-cluster engagement, functioning as an emotionally driven advocacy community. Its high internal cohesion (85% of activities occurring within the cluster) demonstrates strong group solidarity while maintaining the highest outreach rate compared to other clusters. This dual characteristic suggests a confident community with a strong internal identity and proactive messaging strategy.

In contrast, the Counter Mario Dandy cluster exhibited a more balanced engagement pattern, with 25% of its activities directed toward external communities. This increased porosity in boundaries reflects the cluster’s analytical approach, which prioritized broader contextual understanding over mere advocacy. Its significant interaction with the Media cluster (5594 interactions) indicates a strategy focused on factual validation and context-building, distinguishing them from the more emotionally driven Pro David cluster.

The News Media cluster, despite its relatively smaller size, played a crucial mediating role in the network ecosystem. Its balanced interaction distribution facilitated the flow of information across otherwise polarized clusters, acting as what network theory describes as “bridge nodes” [29]. This position allowed it to maintain journalistic neutrality while enabling cross-community dialogue, highlighting the unique role of media actors in digital discourse environments.

Perhaps most revealing was the behavior of the Buzzer cluster, which, despite accounting for only 6.05% of accounts, generated disproportionate activity (3.54 actions per account versus the network average of 3.18). Its limited substantive contribution (only 5% of direct interactions) but high visibility exemplify what Castells [12] describes as “noise generation” in digital networks—high-volume, low-impact communication that occupies space without advancing meaningful discourse.

The interplay between these clusters, particularly the dynamic tension between victim-centered and context-oriented communities, shaped how the case was understood and discussed in the public sphere. While digital discourse is often characterized as highly polarized, our finding that 20% of total activities involved cross-cluster interaction challenges assumptions about absolute echo chambers in social media discourse on sensitive topics.

3.3. Information Flow and Interaction Patterns

The Pro David cluster (refer to Figure 3), which comprised the largest portion of users (40.12% with 33,338 accounts), demonstrated a highly focused and emotionally charged narrative structure. Analysis of the word cloud reveals several dominant themes and narrative patterns that characterized this cluster’s discourse.

• Pro David cluster

The network visualization, centered around influential accounts like “@logikapolitikid”, shows how information and narratives flowed within this cluster. The dense interconnections visible in the blue-tinted section demonstrate the cluster’s high internal cohesion (85% internal activities), while the extending tendrils indicate how these narratives spread to other clusters.

The cluster’s discourse was characterized by four main narrative threads:

• Focus on Mario’s violent actions toward David, emphasizing the brutality of the incident.
• Strong advocacy for severe legal punishment, specifically calling for extended imprisonment.
• Expansion of accountability to include other parties, particularly demanding police action against Agnes.
• Support for the victim’s family’s legal position, specifically their request for attempted murder charges.

The most prominent terms in the visualization (refer to Figure 4)—“mario”, “dandy”, and “david”—indicate the central focus of the discourse, but it is the surrounding terms that reveal the cluster’s emotional and argumentative framework. Terms like “penganiayaan” (assault), “keadilan” (justice), and “menyesal” (regret) suggest a strong emphasis on the criminal nature of the incident and demands for accountability.

The interaction dynamics of the Pro David cluster reveal sophisticated patterns of information exchange and strategic communication behaviors. With 8334 interactions (25% of their external activities) directed towards the News Media cluster, this group demonstrated a strong reliance on mainstream media for information validation and narrative amplification. This high level of media engagement suggests a deliberate strategy to strengthen the group’s position through association with credible news sources, while simultaneously ensuring that its perspective reached broader audiences through established media channels.

The cluster’s engagement with the Mario Dandy cluster, though lower at 5000 interactions (15%), represents a significant level of cross-narrative dialogue. This interaction pattern is particularly noteworthy given the potentially antagonistic nature of the groups’ positions. The lower interaction rate with this opposing cluster, compared to media engagement, suggests a strategic choice to prioritize message amplification through credible channels rather than direct confrontation with opposing viewpoints.

The high message amplification rate within the cluster deserves special attention. This characteristic manifests through the following:

• Rapid sharing of news articles supporting their narrative.
• Consistent amplification of victim-supportive content.
• Coordinated hashtag usage to maintain topic visibility.
• Strategic redistribution of media coverage aligning with their perspective.

These interaction patterns reveal a sophisticated understanding of digital information flows, where the cluster effectively balanced three key activities:

• Validating their narrative through media engagement.
• Maintaining selective dialogue with opposing viewpoints.
• Amplifying their message through internal network activation.

The strategic distribution of interactions (25% media, 15% opposition, 60% internal) suggests a well-organized approach to maintaining narrative control while ensuring their perspective remained prominent in the broader public discourse.

The discourse dynamics in the Pro David cluster are best exemplified by influential accounts that shaped public opinion through detailed case information, vigilant monitoring, and emotional framing. A representative example comes from @LenteraBangsaa_, an influential account that provided specific details: “Korban atas nama David dan pelaku utama bernama Mario Dandy Satriyo menggunakan kendaraan Rubicon 120 den (plat aslinya 2571 pbp). Pelaku utama merupakan lulusan Taruna Nusantara.” [The victim named David and the main perpetrator named Mario Dandy Satriyo used a Rubicon vehicle 120 den (original plate 2571 pbp). The main perpetrator is a Taruna Nusantara graduate.] This tweet demonstrates how this cluster maintained detailed documentation of the case, including vehicle identification and the educational background of the perpetrator—emphasizing evidence collection and fact-based advocacy for the victim.

The cluster’s vigilance in monitoring the case is evidenced by tweets using hashtags like #KawalDavid [#GuardDavid]: “Besok sidang pemeriksaan saksi ke-4 untuk berkas perkara Mario Dandy dan Shane Lukas; anak AG yang sudah inkrach putusannya juga akan memberikan kesaksiannya besok. Beberapa hal yang kita harap akan diterangkan oleh saksi anak ini termasuk yang semestinya digali oleh JPU:” [Tomorrow is the fourth witness examination trial for Mario Dandy and Shane Lukas’ case files; AG’s child whose verdict is already final will also give testimony tomorrow. Several things we hope will be explained by this child witness, including those that should be explored by prosecutors:]. This demonstrates the cluster’s commitment to sustained advocacy throughout the legal process.

Influential opinion leader @logikapolitikid used satire to critique perceived preferential treatment: “El, gantiin penyidik Polres Jaksel bentar aja, soalnya kasus si Mario Dandy ini udah bau-bau ‘aroma Sambo’ kayaknya. Atau gantiin kapolresnya sekalian..😁” [El, replace the South Jakarta Police investigator just for a moment, because this Mario Dandy case already smells like the “Sambo aroma” it seems. Or replace the police chief altogether..😁]. This comparison to the high-profile Sambo case (where a police general murdered his aide) suggests suspicion of preferential treatment for elite perpetrators—a recurring theme in this cluster’s discourse.

The cluster also actively challenged media narratives, as demonstrated by @seeksixsuck: “Masih inget nggak, pada hari semua media posting: Mario Dandy tidak pernah ditengok keluarga. Bohong besar!!! Bapaknya Shane cerita kalau ketemu ortu Dandy di tahanan Polda Metro sama-sama bezuk. Dan dia dicuekin ortu Dandy. Mafia ini membeli media dengan nilai fantastis” [Remember when all media posted: Mario Dandy was never visited by family. Big lie!!! Shane’s father tells that he met Dandy’s parents at the Metro Police detention during visits. And he was ignored by Dandy’s parents. This mafia buys media for fantastic sums]. This tweet exemplifies how the cluster positioned itself as a counter-narrative force against perceived manipulation, consistently advocating for truth and justice for the victim while challenging what they viewed as elite privilege influencing the media narrative.

2.

Counter Mario Dandy cluster

The Counter Mario Dandy cluster (refer to Figure 5), representing 26.93% of total users, with 23,314 accounts, demonstrated a distinct narrative focus that went beyond the immediate incident to examine broader societal implications. Analysis of the cluster’s 1999 tweets reveals sophisticated patterns of discourse that emphasized structural and contextual aspects of the case.

The network visualization, centered around key accounts like “@logikapolitikid” and “@detikcom”, shows how this cluster maintained a more balanced interaction pattern:

• 75% internal activities demonstrating strong group coherence.
• 25% external engagement showing openness to broader dialogue.
• Significant connections to media nodes indicating active fact-checking and context-building.

This cluster’s approach was characterized by the following:

• Comparative analysis with other criminal cases.
• Focus on Mario Dandy’s perceived arrogance and privilege.
• Broader discussions about social inequality.
• Critical examination of institutional responses.

The relatively high ratio of accounts (23,314) to tweets (1999) suggests that while the cluster had significant support, its core messaging was more focused and coordinated compared to other clusters.

This analysis reveals how the cluster effectively broadened the discourse from a single incident to a wider critique of social privilege and institutional accountability in Indonesian society.

The word cloud visualization reveals several key narrative threads (refer to Figure 6):

• The prominence of terms like “pejabat” (official), “pajak” (tax), and “kasus” (case) indicates a focus on the socio-economic context and privilege aspects of the case.
• Terms like “penganiayaan” (assault) appear alongside “hukum” (law), suggesting an emphasis on legal accountability.
• The presence of words like “warga” (citizen) and “anak” (child) points to discussions about social class disparities.
• References to “bapaknya” (his father) and “anaknya” (his child) show attention to family dynamics and parental responsibility.

The interaction dynamics of this cluster reveal a more balanced and deliberate approach to information exchange and discourse participation. With 5594 interactions (17%) directed toward the News Media cluster, this group demonstrated a strategic approach to information validation and context-building. Unlike the Pro David cluster’s higher media engagement (25%), this more moderate level of media interaction suggests a different strategy—one focused on using media sources selectively to support broader social commentary rather than purely amplifying news coverage.

The cluster’s engagement with the Pro David cluster, accounting for 4475 interactions (13%), demonstrates an interesting pattern of cross-narrative dialogue. This relatively balanced ratio between media interactions (17%) and engagement with opposing viewpoints (13%) reflects the cluster’s more evenly distributed communication strategy. This pattern suggests a conscious effort to achieve the following:

• Validate information through media sources.
• Maintain constructive dialogue with opposing perspectives.
• Build broader context around the case.
• Engage in substantive discussion rather than mere opposition.

The more balanced interaction distribution is particularly noteworthy because it suggests the following:

• A more sophisticated approach to discourse participation.
• Greater willingness to engage with diverse perspectives.
• Strategic use of both media sources and direct dialogue.
• Focus on contextual understanding rather than a purely emotional response.

This balanced interaction pattern aligns with the cluster’s broader narrative focus on structural issues and social context, showing how the group’s communication strategy supported its goal of expanding the discussion beyond the immediate incident to broader societal implications. The relatively even distribution of external interactions (17% media, 13% opposition) suggests a deliberate strategy to maintain credibility while engaging in substantive dialogue across different perspectives in the network.

The Counter Mario cluster’s approach to contextualizing the case beyond the immediate incident is clearly demonstrated by influential accounts that connect individual violence to broader societal issues and emphasize legal accountability. A significant example comes from @mazzini_gsp, a prominent opinion leader who focused on the legal dimensions: “polisi menjerat mario dandy dengan sangkaan lebih berat yakni pasal 355 kuhp ayat tentang penganiayaan berat berencana, dengan ancaman hukuman maksimal 12 tahun”. [Police charged Mario Dandy with a more serious accusation under Article 355 of the Criminal Code concerning premeditated severe assault, with a maximum penalty of 12 years.] This tweet reflects the cluster’s emphasis on the judicial process and advocacy for appropriate legal consequences rather than purely emotional responses.

The cluster’s tendency to connect individual cases to broader patterns of problematic behavior among privileged youth is exemplified by @wgs1996’s’s comment: “banyak anak muda seumuran mario dandy, anak akbp achiruddin dan satu ini yg bertingkah jauh dari norma. ini adalah bukti kegagalan pendidikan moral oleh orangtua dari satu generasi yg sama. 🫠” [Many young people of Mario Dandy’s age, AKBP Achiruddin’s child and this one behave far from norms. This is evidence of a failure in moral education by parents from the same generation.] This tweet demonstrates how this cluster consistently frames youth violence as a systemic social issue rather than isolated incidents, focusing on patterns and root causes in parenting and moral education.

The temporal commitment to case monitoring characteristic of this cluster is evidenced by @eradotid’s’s tweet: “Btw, di sidang perdana ini, si Mario Dandy sama Shane Lukas didakwa pasal penganiayaan berencana dan terancam hukuman 12 tahun penjara. Kita kawal terossss pokonya!!!✊” [Btw, in this initial trial, Mario Dandy and Shane Lukas are charged with premeditated assault and face 12 years in prison. We’ll keep monitoring for sure!!!✊]. This demonstrates the cluster’s advocacy for sustained public attention throughout the legal process, focusing on systemic accountability rather than immediate emotional reactions.

Another dimension of this cluster’s discourse is reflected in @BISMOLICIOUS’s comment in response to @asumsico: “Klo ada permintaan dari Tuhan untuk Mario Dandy cs, yakin banget permintaan mereka cuma pengen waktu diulang kembali dan ga ngelakuin hal-hal tsb ke David.. yakin banget... awalnya ngerasa jumawa dan di atas.. skrg jatuh se jatuh-jatuhnya ke dasar bumi... semua tampak kacau dan kelam🤣” [If there’s a request from God for Mario Dandy and friends, I’m certain their only request would be to turn back time and not do those things to David... very certain... initially feeling arrogant and superior... now falling to rock bottom... everything seems chaotic and dark🤣]. This tweet illustrates the cluster’s moral framing, emphasizing consequences and reflection rather than pure condemnation, a more measured approach that analyzes behavior patterns and their outcomes.

Comparison with similar cases was another common strategy in this cluster, as shown by @lillbunnybunn’s observation: “bjir yang kaya begini bocah baru masuk sma? berasa deja vu kaya kasusnya mario dandy. orang kuliahan pacaran sama bocah smp berujung bawa sial”. [Wow, someone like this just entered high school? Feels like deja vu like Mario Dandy’s case. College students dating middle schoolers ending up bringing bad luck.] This tweet demonstrates how the cluster positioned the Mario Dandy case within a broader pattern of age-inappropriate relationships and power imbalances, emphasizing systemic social issues rather than focusing solely on individual blame.

3.

News Media cluster

The News Media cluster (refer to Figure 7), comprising 14.14% of total engagement, with 10,559 accounts generating 10,910 tweets, demonstrates a highly structured and professional approach to information dissemination. The nearly 1:1 ratio between accounts and tweets (10,559:10,910) suggests consistent, measured reporting rather than rapid-fire information sharing typical of other clusters.

The network visualization, particularly around nodes like “@detikcom”, shows the media cluster’s central position in information flow:

• Strategic positioning between opposing clusters.
• Dense connections indicating high content adoption rates.
• Balanced distribution of outgoing connections suggesting a neutral stance.

The cluster’s professional reporting approach is evidenced by the following:

• Consistent output (average 1.03 tweets per account).
• Focus on development updates.
• Balanced coverage of all involved parties.
• Emphasis on factual verification.

This analysis reveals how the Media cluster maintained its role as a primary information source while serving as a bridge between different narrative perspectives in the broader network.

Analysis of the word cloud reveals several key characteristics of media coverage (refer to Figure 8):

• Factual Focus
  • Prominence of terms like “kasus” (case) and “tersangka” (suspect) indicates objective reporting.
  • Use of “penganiayaan” (assault) as a formal legal term rather than emotional descriptors.
  • Integration of official terms like “pejabat” (official) and “dirjen” (director general).
• Comprehensive Coverage
  • Multiple actor references: “mario”, “dandy”, “satriyo”, “agnes”.
  • Legal aspects: “hukum” (law), “polisi” (police).
  • Context elements: “video”, “viral”, “saksi” (witness).

The News Media cluster’s 7049 cross-cluster interactions reveal a sophisticated and balanced approach to information dissemination. Unlike other clusters that showed clear interaction preferences, the Media cluster maintained remarkably balanced distribution patterns across major clusters, demonstrating these users’ role as neutral information brokers in the network.

This balanced interaction pattern manifested in several key ways:

• Distribution Balance
  • 3525 interactions (11%) with Pro David cluster.
  • 2937 interactions (9%) with Mario Dandy cluster.
  • Remaining interactions distributed among smaller clusters and general audience.
• Content Reliability Metrics
  • Highest verification rates for shared information.
  • Consistent fact-checking practices.
  • Regular updates with verified sources.
  • Strong correlation between reported events and official statements.

• The cluster’s high content reliability metrics were demonstrated through the following:
  • Systematic source verification.
  • Consistent fact-checking protocols.
  • Regular updates with official statements.
  • Clear distinction between facts and allegations.

This balanced interaction pattern, combined with high reliability metrics, suggests that the Media cluster effectively fulfilled its role as follows:

• Primary information validator.
• Neutral information broker.
• Bridge between opposing narratives.
• Reliable source for all stakeholders.

The data indicate that while other clusters might have generated more total interactions, the Media cluster’s balanced approach and high reliability made it a crucial node in the network’s information ecosystem, facilitating fact-based discourse across different perspectives.

The balanced, information-focused approach characteristic of the Media cluster is exemplified by mainstream media accounts that maintained neutral reporting while providing comprehensive coverage of case developments. @CNNIndonesia exemplified this with their tweet: “Mantan pegawai DJP Kemenkeu Rafael Alun Trisambodo mengaku jadi terseret kasus hukum buntut kekerasan yang dilakukan anaknya, Mario Dandy Satriyo. Rafael mengklaim sedang dicari-cari kesalahannya dari harta yang ia miliki. #CNNIndonesia” [Former DGT Ministry of Finance employee Rafael Alun Trisambodo admits to being drawn into legal cases following violence committed by his son, Mario Dandy Satriyo. Rafael claims his wealth is being scrutinized for mistakes. #CNNIndonesia]. This tweet, which received 347 shares, demonstrates the cluster’s commitment to reporting statements from all parties involved while maintaining a journalistic distance through phrases like “mengklaim” (claims) that subtly indicate the subjective nature of the source’s statement.

The News Media cluster’s role in documenting significant trial developments is evident in @KompasTV’s coverage: “Terdakwa Mario Dandy Satriyo divonis 12 tahun penjara dan restitusi sebesar Rp25 miliar atas kasus penganiayaan David Ozora pada hari ini (/) di Pengadilan Negeri Jakarta Selatan. Simak berita selengkapnya di... #SidangMarioDandy” [Defendant Mario Dandy Satriyo was sentenced to 12 years in prison and restitution of Rp25 billion for the David Ozora assault case today (/) at the South Jakarta District Court. See the complete news at... #MarioDandyTrial]. This factual reporting of judicial outcomes without editorial commentary typifies the cluster’s information-centered approach.

The ongoing nature of media coverage is demonstrated by @kumparan’s updates throughout different phases of the case: “Mario Dandy Satrio (20) mengaku cemas dengan keadaan David Ozora (17), yang dia aniaya di kawasan Pesanggrahan, Jakarta Selatan beberapa waktu lalu. Kekhawatiran Mario itu disampaikan oleh kuasa hukumnya, Dolfie Rompas. #update” [Mario Dandy Satrio (20) admits concern about the condition of David Ozora (17), whom he assaulted in the Pesanggrahan area, South Jakarta some time ago. Mario’s concern was conveyed by his attorney, Dolfie Rompas. #update]. This tweet demonstrates how media outlets consistently reported developments while maintaining factual accuracy in attribution (“disampaikan oleh kuasa hukumnya”) without evaluative language.

The cluster’s comprehensive coverage approach is evident in @detikcom’s detailed thread format: “Perempuan inisial AG (15), pelaku anak di kasus Mario Dandy Satriyo (20) menganiaya Cristalino David Ozora (17), akhirnya resmi ditahan. Penahanan tersebut dilakukan setelah ia diperiksa selama jam sebagai pelaku anak. Simak informasi selengkapnya dalam utas berikut. #thread” [Female with initials AG (15), juvenile perpetrator in the case of Mario Dandy Satriyo (20) assaulting Cristalino David Ozora (17), has finally been officially detained. The detention was carried out after she was examined for hours as a juvenile perpetrator. See more information in the following thread. #thread]. This demonstrates the cluster’s commitment to providing contextual details (ages of all involved parties, procedural information) while using neutral language.

While media accounts maintained journalistic neutrality, comments on their posts often reflected strong public sentiment, as seen in @aku_saja87’s response to @kumparan: “Pasti dia anaknya orng miskin gak berduit. Coba saja berduit seperti Mario Dandy pasti di jamin aman ama polisi” [He must be a child of poor people without money. If only he had money like Mario Dandy, he would surely be guaranteed safety by the police]. Similarly, @baperanewscom reported “Mario Dandy & kuasa hukum terlihat tertawa saat ayah David menegaskan bahwa anaknya David belum bisa mandi & belum bisa pakai celana. #MarioDandy” [Mario Dandy & his lawyer were seen laughing when David’s father affirmed that his son David still cannot bathe & cannot wear pants]. These examples illustrate how media reporting often formed platforms for public discourse while the outlets themselves maintained professional distance in their presentation.

4.

Political Buzzer cluster

The Political Buzzer cluster (refer to Figure 9), while representing only 6.05% of total users, demonstrates a distinct pattern of opportunistic engagement with the Mario Dandy case. Analysis of the word cloud reveals a striking departure from case-related discourse, showing how this cluster attempted to leverage the case’s visibility for different agendas.

The word cloud analysis reveals several key characteristics (refer to Figure 10):

• Political Context Integration
  • Presence of terms like “jokowi” and “milenial” indicates attempts to politicize the case.
  • Location-specific terms (“cilacap”, “bromo”) suggest regional political messaging.
  • Terms unrelated to the case (“ariani”, “yunanda”) show attempts to piggyback on trending topics.
• Divergent Narrative Focus
  • Limited use of case-specific terms.
  • Introduction of unrelated political and social terms.
  • Inclusion of entertainment and celebrity references.

• The network visualization shows this cluster’s peripheral position:
  • Isolated position in the network structure.
  • Limited meaningful connections to main discourse clusters.
  • Sporadic bursts of activity rather than sustained engagement.

The cluster’s behavior is characterized by the following:

• Opportunistic hashtag usage.
• High activity per account (3.54).
• Limited substantive contribution (5% of direct interactions).
• Focus on visibility rather than meaningful discourse.

This analysis reveals how members of the Buzzer cluster operated as digital opportunists, attempting to capitalize on the case’s visibility while contributing minimal substantive content to the actual discussion. Their positioning and behavior patterns suggest a strategic approach focused on visibility and reach rather than meaningful engagement with the case’s core issues.

The Political Buzzer cluster, despite its relatively small size of 5027 accounts (6.05% of total users), demonstrated distinctive behavioral patterns that set it apart from other clusters. Most notably, this cluster recorded the highest activity per account ratio, at 3.54, significantly exceeding the network average, suggesting intensive but potentially automated or coordinated activity patterns.

The cluster’s interaction patterns reveal a strategic but superficial engagement approach:

• Cross-cluster Interactions
  • Only 2593 interactions representing 5% of total cross-cluster activities.
  • Limited meaningful dialogue with main discussion clusters.
  • Sporadic rather than sustained engagement patterns.
• Content Strategy
  • High retweet ratio with minimal original content generation.
  • Strategic hashtag deployment targeting visibility metrics.
  • Opportunistic topic exploitation without substantive contribution.
  • Focus on amplification rather than content creation.

The behavioral metrics indicate a distinct operational pattern:

• Engagement Characteristics
  • Short-burst, high-volume activity periods.
  • Limited sustained presence in any single discussion thread.
  • Strategic timing of posts to maximize visibility.
  • Minimal investment in meaningful discourse.
• Network Position
  • Peripheral location in the overall network structure.
  • Weak connections to main discussion clusters.
  • Limited influence on core narrative development.
  • Focus on volume rather than impact.

These patterns suggest a cluster primarily focused on exploiting the case’s visibility for alternative agendas rather than contributing to meaningful discussion about the youth violence issue. Their high activity ratio combined with limited substantive engagement reveals a strategic approach prioritizing visibility over content value.

The data indicate that while this cluster maintained high activity levels, its contribution to the broader discourse remained minimal, with members functioning more as digital opportunists than genuine discussion participants.

3.4. Temporal Evolution and Activity Patterns

The analysis revealed distinct temporal dynamics across different clusters:

• Daily Activity Distribution
  • Pro David cluster: Peak activity 14:00–17:00 WIB (35% of activities);
  • Mario Dandy cluster: Peak 19:00–22:00 WIB (40% of activities);
  • News Media cluster: Highest activity 08:00–11:00 WIB (45% of activities);
  • Buzzer cluster: Sporadic activity peaks (not time-bound).
    • Multiple short bursts of high activity;
    • No consistent temporal pattern;
    • Opportunistic engagement with trending moments.
• Content Lifecycle Phases
  • Initial Phase (0–24 h)
    • 60% engagement focused on factual details;
    • Breaking news reached 800 engagements;
    • Decay rate −15% per hour.
  • Middle Phase (24–72 h)
    • More stable engagement (300–450 per post);
    • Moderate decay rate (−8% per hour);
    • Focus shifted to context and analysis.
  • Advanced Phase (>72 h)
    • Lower but sustained engagement (200–300 per post);
    • Extended discussion period up to 48 h;
    • Focus on systemic issues.

The temporal analysis of discourse evolution revealed distinct phases that align with and extend theoretical frameworks on digital communication dynamics. Far from being a static phenomenon, the discourse surrounding the Mario Dandy case demonstrated clear evolutionary patterns that reflect what van Dijck et al. [14] describe as the “platform society” temporal logic—the acceleration, amplification, and eventual transformation of public discourse through digital media mechanisms.

In the initial phase (0–24 h), we observed communication patterns dominated by information dissemination rather than dialogue, with 60% of engagement focused on factual details. This aligns with Castells’ [12] concept of “viral politics”, where the urgency of information sharing initially supersedes reflective dialogue. Breaking news content during this phase reached peak engagement of approximately 800 interactions per post, with the Media cluster exercising significant influence as primary information providers. This influence was evident in their activity concentration during morning hours (08:00–11:00 WIB), accounting for 45% of their total activities and establishing the initial factual framework that other clusters would subsequently interpret.

The transition to the middle phase (24–72 h) demonstrated what Monge and Contractor [29] theorize as the “equilibration phase” in network communication, where initial information saturation gives way to more stable engagement patterns. During this period, engagement stabilized at 300–450 interactions per post, with a moderate decay rate of 8% per hour—significantly lower than the initial phase’s 15% hourly decay. More importantly, the discourse underwent a qualitative transformation from predominantly factual reporting to contextual analysis and interpretation. The Pro David cluster became most active during this phase, with peak activity concentrated between 14:00 and 17:00 WIB (35% of their activities), suggesting strategic timing aimed at maximum audience reach during late afternoon hours.

The advanced phase (>72 h and beyond) revealed the most theoretically significant evolution—what Boyd [13] describes as the shift from “networked publics” to “networked counterpublics”. In this phase, discourse transcended the specific case to address broader systemic issues, with the Mario Dandy cluster becoming most influential as peak activity shifted to evening hours (19:00–22:00 WIB, 40% of their activities). This timing suggests a more deliberative environment where users engaged in longer, more reflective interactions. Engagement during this phase, while lower in volume (200–300 per post), demonstrated remarkable sustainability, with extended discussion periods lasting up to 48 h—a pattern that challenges assumptions about digital discourse being inherently ephemeral.

These temporal patterns extend Wardle and Derakhshan’s [27] Information Disorder Framework by demonstrating how different types of information (misinformation, disinformation, and malinformation) manifest and evolve distinctly across temporal phases. In the initial phase, misinformation dominated as users shared unverified details, while the advanced phase saw more sophisticated disinformation as political opportunists attempted to leverage the established narrative for unrelated agendas. This temporal evolution provides a more nuanced understanding of how digital discourse transforms not only in volume and reach but also in fundamental character and function across its lifecycle.

3.5. Content Impact and Engagement Patterns

Analysis of content impact revealed three distinct categories:

• High-Impact Content (>1000 engagements)
  • Average lifespan: 48 h;
  • Peak engagement: 4–6 h post-publication;
  • Distribution: Pro David (45%), Mario Dandy (30%), Media (20%);
  • Sustained significant engagement up to 24 h.
• Medium-Impact Content (100–1000 engagements)
  • Average lifespan: 24 h;
  • Peak engagement: 2–4 h post-publication;
  • Distribution: Pro David (40%), Mario Dandy (35%), Media (15%);
  • More balanced cluster distribution.
• Response Time and Engagement
  • Media content: Fastest response (<30 min);
  • Pro David content: Initial pickup ~45 min;
  • Mario Dandy content: Initial pickup ~60 min;
  • Political Buzzer content: Quick initial response (<15 min);
  • Sustained engagement periods: 3–5 h average.

The story of how content shaped public perception of the Mario Dandy case unfolds through three distinct content categories, each playing unique roles in the evolving digital narrative. What began as shock and outrage transformed into systemic critique through a complex interplay of content types and engagement patterns that reveal much about how Indonesian digital society processes cases of youth violence.

High-impact content—posts exceeding 1000 engagements—functioned as narrative anchors in the discourse ecosystem. These powerful communications, comprising just 3% of the total content but capturing 41% of all engagement, established the dominant framing that subsequent discussions would either support or challenge. The distribution of these narrative anchors revealed significant power imbalances across clusters: the Pro David cluster produced 45% of these viral posts, while the Mario Dandy cluster accounted for 30%, and mainstream media generated 20%. This asymmetrical distribution demonstrates how emotional framing centered on victim advocacy captured a greater immediate public response than contextual analysis or factual reporting.

The journey of high-impact content followed a consistent lifecycle that illuminates digital attention dynamics. These posts typically reached peak engagement 4–6 h after publication—notably during evening hours (19:00–22:00 WIB)—and maintained significant traction for approximately 24 h. One particularly illustrative example came from an influential account in the Pro David cluster, whose emotional appeal for justice received over 5600 shares and spawned multiple trending hashtags. The post’s simple yet powerful framing—positioning the case as emblematic of elite impunity—resonated deeply across demographic boundaries, demonstrating how moral framing that connects individual incidents to systemic issues generates the strongest public response.

Medium-impact content (100–1000 engagements) served a different but equally vital function in the narrative ecosystem. These posts, representing 22% of content but 37% of total engagement, operated as connective tissue between viral narratives and specialized discussions. Their more balanced distribution across clusters (40% Pro David, 35% Mario Dandy, 15% Media) created bridges between competing interpretations, allowing for a more nuanced understanding than high-impact content alone could provide. The shorter lifecycle of these posts—peaking at 2–4 h and sustaining significant engagement for approximately 12 h—contributed to a continuous flow of fresh perspectives that prevented discourse stagnation while maintaining thematic consistency.

Perhaps most revealing was the relationship between response time and sustained engagement. Media content received the fastest initial response (under 30 min), reflecting public hunger for authoritative information, while Pro David content (45 min average initial response) and Mario Dandy content (60 min average) followed behind. However, this pattern reversed for sustained engagement, with emotionally resonant Pro David content maintaining conversation threads for an average of 5 h compared to 3 h for more factual media content. This pattern illuminates the tension between information hunger and emotional resonance that characterizes digital discourse around traumatic events.

These engagement patterns tell a story beyond mere metrics—they reveal how Indonesian digital society processes complex social traumas through collective sense-making. The evolution from shock (rapid engagement with factual content) to emotional response (sustained engagement with advocacy content) to reflective analysis (extended discussion of systemic implications) demonstrates a sophisticated collective processing that challenges simplistic views of social media discourse as merely reactive or superficial.

3.6. Network Resilience and Sustainability

The network demonstrated strong resilience characteristics:

• Structural Resilience
  • Path reliability: 85% success rate;
  • Average path length: 2.3 hops;
  • Secondary paths: 15,849 backup routes;
  • Multiple connection pathways between major clusters.
• Connection Distribution
  • Strong ties: 25% (20,774 connections);
  • Medium ties: 45% (37,394 connections);
  • Weak ties: 30% (24,929 connections).
• Sustainability Metrics
  • Core user retention: 65%;
  • Content type distribution:
    • Evergreen content: 15%;
    • Time-sensitive content: 70%;
    • Ephemeral content: 15%.
• Network Impact
  • Direct activities: 264,155;
  • Secondary views: 792,465;
  • Total impressions: 1,584,930;
  • Multiplier effect: six impressions per interaction.

The Mario Dandy discourse network demonstrated remarkable resilience and sustainability, challenging conventional understanding of digital conversation lifecycles. Rather than following the typical pattern of rapid amplification followed by quick decay, this network exhibited structural characteristics that enabled prolonged engagement and evolving discourse over several months. This resilience reveals important insights into how digital networks maintain momentum around socially significant issues.

The network’s structural resilience was evident in its robust connectivity metrics. With a path reliability of 85%, communication could flow through the network even when accounting for potential node removal or inactivity—a characteristic that Monge and Contractor [29] identify as critical for sustained information diffusion in complex networks. The average path length of 2.3 hops indicates remarkable network efficiency, allowing information to travel from any point to another with minimal intermediaries. This efficient structure was supported by an impressive 15,849 backup routes between major clusters, ensuring that no single point of failure could disrupt the overall discourse flow. These redundant pathways created what Castells [12] describes as “self-repairing networks”, where communication patterns can adapt and reconstitute even when individual nodes or connections become inactive.

The distribution of connection types proved particularly significant for understanding network sustainability. Strong ties (representing 25% of connections with 20,774 relationships) provided the stable core that maintained consistent narrative threads over time. Medium ties (45% with 37,394 connections) facilitated information exchange between semi-related communities, while weak ties (30% with 24,929 connections) created pathways for new information and perspectives to enter the discourse. This balanced distribution exemplifies what Granovetter’s [38] seminal work described as the “strength of weak ties” phenomenon, where diverse connection types collectively create a robust communication ecosystem. The network’s sustainability directly contradicts predictions that digital discourse is inherently ephemeral, instead demonstrating how properly structured networks can maintain engagement over extended periods.

The core user retention rate of 65% further illustrates this sustainability—nearly two-thirds of users who engaged during the first week of the case continued participating throughout the observed period. This retention sharply contrasts with typical social media engagement patterns that show rapid participant turnover. The distribution of content types supported this sustained engagement: 15% evergreen content that maintained relevance throughout the case timeline, 70% time-sensitive content tied to specific developments, and 15% ephemeral content addressing momentary aspects. This content ecology created what van Dijck et al. [14] describe as “sustained attention cycles” where new developments continually refresh engagement with core themes.

Perhaps most revealing was the network’s multiplier effect—the capacity to amplify initial engagement into broader impact. The 264,155 direct activities generated an estimated 792,465 secondary views and approximately 1,584,930 total impressions, creating a multiplier effect of roughly six impressions per interaction. This amplification occurred in a consistent pattern of a 1:5:15 (creator:engager:viewer) ratio, demonstrating how the network efficiently propagated content beyond immediate participants to reach broader audiences. This multiplier effect illustrates what Bennett and Segerberg [39] term “connective action” in digital networks, where individual activities collectively generate impact disproportionate to their initial scale.

The bridge nodes—accounts that facilitated cross-cluster dialogue—proved critical for network sustainability. Comprising 10–15% of accounts but facilitating approximately 20% of cross-cluster interactions, these bridges prevented the network from fragmenting into isolated echo chambers. Their consistent activity throughout the case timeline ensured continuous dialogue across ideological and narrative boundaries, maintaining what Benkler [40] describes as “productive friction”, which prevents discourse stagnation. These nodes, predominantly from the Media cluster but also including individual influencers and subject matter experts, maintained critical pathways that allowed the network to evolve rather than calcify into rigid positions.

Together, these characteristics reveal a digital discourse network with sophisticated self-sustaining mechanics that enabled prolonged engagement with complex social issues. Far from the ephemeral “viral moments” that often characterize social media, the Mario Dandy case network demonstrates how digital discourse can maintain momentum, evolve narratively, and sustain public attention when proper structural conditions exist. This resilience has significant implications for understanding how digital public spheres can support sustained engagement with important social issues rather than merely facilitating momentary attention spikes.

4. Discussion

4.1. Key Actors in Narrative Formation and Dissemination

The analysis reveals complex dynamics in how key actors shape and spread narratives in digital public spaces. The Pro David cluster (40.12%) emerged as the dominant force, with high internal cohesion (85% internal activities) demonstrating effective narrative building around victim support. Their strong cross-cluster interactions (15,001 interactions) suggest an active role in shaping broader public discourse. This finding both supports and challenges Castells’ [12] Network Society Theory. While it confirms the shift toward network-based power flows, the high internal cohesion suggests that hierarchical structures persist within network architectures, creating what might be termed “networked hierarchies”.

Members of the Mainstream Media cluster, despite comprising only 14.14% of accounts, played a crucial role as information validators. Their balanced interaction distribution and high content adoption rates demonstrate their function as bridges between different perspectives. This finding aligns with van Dijck et al.’s [14] concept of a platform society, where media institutions adapt to digital platform logic while maintaining their role as primary information sources. However, it also challenges the Platform Society Framework by demonstrating how traditional media institutions maintain a significant influence despite platform disintermediation.

This tension between platform mechanics and institutional authority is particularly evident in the Indonesian context. As Lim [31] argues, cultural factors significantly influence how digital platforms are utilized, creating hybrid forms of authority that combine traditional institutional power with network dynamics. The data support this through the media cluster’s high content reliability metrics and balanced interaction patterns, suggesting that cultural trust in established institutions remains robust even within digital networks.

The emergence of the Political Buzzer cluster (6.05%) presents an interesting challenge to existing theoretical frameworks. While platform society theory predicts the emergence of opportunistic actors, the cluster’s behavior patterns—high activity rates but limited substantive engagement—suggest more complex motivations than simple algorithmic exploitation. This supports the Information Disorder Framework Tomassi et al. [26] and Wardle & Derakhshan [27] while suggesting the need for a more nuanced understanding of how different types of actors navigate digital spaces.

4.2. Information Dissemination Patterns

The dominance of retweets (68%) over dialogic interactions (4% combined for replies, mentions, and quotes) reveals a significant pattern in how information spreads in Indonesian digital spaces. This “share first, discuss later” behavior suggests that information amplification takes precedence over substantive discussion, reflecting what Boyd [13] describes as networked publics’ characteristics. However, this pattern presents an interesting challenge to traditional communication theory assumptions about information diffusion and public discourse formation.

The high proportion of retweets raises important theoretical questions about the quality of digital public discourse. While Castells [12] emphasizes the democratizing potential of network communication, the low level of dialogic interaction might suggest what Fuchs [41] terms “pseudo-participation”—high activity levels masking limited substantive engagement. Yet, the data reveal a more complex picture: the significant cross-cluster interactions (20% of total activities) challenge common assumptions about echo chambers on social media.

This finding particularly contests van der Linden et al.’s [42] arguments about digital polarization. The network demonstrated substantial information flows between different opinion clusters, particularly through bridge nodes that facilitated cross-community dialogue. This suggests that Indonesian digital discourse, at least in high-profile youth violence cases, maintains some degree of inter-community dialogue despite clear opinion polarization. This observation aligns more closely with Lim [31] understanding of how cultural factors influence digital communication patterns in Southeast Asian contexts.

The temporal evolution of information flows presents another theoretical challenge. The identification of distinct phases in content lifecycle—from initial amplification (60% engagement with factual content) to later interpretative discussion—suggests a more sophisticated pattern than simple viral spread. This finding extends Monge and Contractor’s [29] network diffusion models by demonstrating how different types of content follow different dissemination patterns at different stages of public discourse.

Building on platform society concepts, Stray et al. [15] suggest the need for more proactive and long-term approaches to managing digital conflict, operating through platform design rather than solely relying on content moderation policies. This finding supports but also complicates van Dijck et al.’s [14] framework by demonstrating how human agency interacts with platform mechanics to shape information dissemination patterns.

The dominance of retweets (68%) over dialogic interactions (4%) observed in our study raises important questions about how platform affordances and content moderation systems shape information flows. As Gosztonyi [35] notes, content moderation decisions by platforms often prioritize easily identifiable violative content rather than addressing more complex narrative patterns. This may partially explain the prevalence of information amplification over substantive dialogue in our dataset.

The high percentage of cross-cluster interactions (20% of total activities) suggests that despite content moderation systems that might filter certain types of speech, Indonesian social media users maintain significant cross-community dialogue. This challenges what Gosztonyi [35] describes as the “filter bubble” effect often attributed to platform algorithms and moderation systems. However, the Political Buzzer cluster’s behavior exemplifies what Gosztonyi terms “organized information operations” that exploit gaps in moderation systems to amplify particular narratives.

4.3. Formation and Evolution of Opinion Clusters

The evolution of opinion clusters demonstrates a complex interplay between platform mechanics, user behavior, and cultural context that both supports and challenges existing theoretical frameworks. The distinct phases identified in cluster evolution present an interesting counterpoint to traditional theories of public opinion formation.

Initial Phase (0–24 h): The dominance of factual reporting (60% engagement) in this phase appears to support van Dijck et al.’s [14] platform society concept, where algorithmic prioritization favors breaking news. However, the rapid cluster formation around initial positions challenges simplistic platform determinism. As Lewandowsky et al. [43] argue, the speed of digital opinion formation can lead to entrenched positions, yet our data show more fluid dynamics:

• Fast initial clustering but with permeable boundaries.
• High information velocity but with significant cross-verification.
• Rapid narrative formation but with multiple competing frames.

Middle Phase (24–72 h): The shift toward interpretative content during this phase presents an interesting challenge to Boyd’s [6] Networked Publics Framework. While her theory emphasizes persistence and replicability, our data show active reframing and recontextualization:

• Stabilization of cluster boundaries while maintaining cross-cluster dialogue.
• Emergence of more nuanced narratives.
• Integration of multiple information sources.
• Development of sustained inter-cluster debates.

Advanced Phase (>72 h): The evolution toward systemic discourse challenges assumptions about the superficiality of social media discussions. This phase demonstrates what Castells [12] terms “programmed networks”, but with important modifications:

• Development of sophisticated analytical frames.
• Integration of multiple perspectives.
• Sustained engagement with deeper societal implications.
• Formation of stable but permeable discourse communities.

The data particularly challenge individualism-centric assumptions about digital polarization. While Sunstein’s [44] theory of echo chambers predicts increasing polarization over time, our findings show the following:

• Increasing cross-cluster dialogue (20% of total activities).
• Development of shared interpretative frameworks.
• Emergence of bridge nodes facilitating inter-group communication.
• Integration of diverse narrative perspectives.

This observation particularly resonates with Lim’s [31] findings about Indonesian digital communication patterns, while extending them to show how cultural factors influence not just initial clustering but the entire evolution of digital discourse. These findings also have important implications for the Information Disorder Framework Tomassi et al. [26] Wardle & Derakhshan [27]. The evolution from factual reporting to systemic analysis suggests that information disorder is not simply about truth versus falsehood, but about how communities collectively make sense of complex social issues through digital interaction.

Our analysis reveals how cluster formation and interaction patterns reflect complex processes of both narrative development and participatory identity construction in digital spaces. The evolution of four distinct clusters transcends simplistic platform dynamics, demonstrating how users actively shape discourse ecosystems through collective agency.

Wardle [28] and Gowder [34] offer complementary frameworks for understanding these phenomena. While Wardle emphasizes narrative accumulation processes, where individual posts gain meaning through their collective contribution to larger stories, Gowder focuses on how these narratives constitute “participatory platform identities” that reflect community values and interests. Both perspectives recognize that digital discourse is not merely algorithmic output but emerges from complex social processes.

The Pro David cluster (40.12%), with its strong narrative consistency supporting the victim, exemplifies both Wardle’s concept of cumulative narrative influence and Gowder’s notion of community-based identity formation. Despite representing a specific position, this cluster maintained significant cross-cluster engagement (15,001 interactions), challenging platform tendencies toward homogenization. This dual characteristic—strong internal coherence with active external engagement—demonstrates what Wardle [28] calls “networked narratives” and what Gowder [34] terms “cross-cultural conflict management” in digital spaces.

The News Media cluster’s unique position further illustrates this theoretical convergence. Functioning as information validators, these accounts (14.14%) embody what Wardle [28] describes as traditional institutions adapting to participatory information ecosystems and what Gowder [34] identifies as “governance entities at multiple scales”. Their balanced interaction patterns across clusters suggest they serve as critical mediators between official narratives and community interpretations, maintaining information integrity while facilitating cross-community dialogue.

Perhaps most significantly, the substantial cross-cluster interactions (20% of total activities) we observed challenge fundamental assumptions in digital media research. These interactions contradict both simplistic echo chamber theories and algorithmic determinism, suggesting that Indonesian users exercise considerable agency in navigating digital spaces. This finding simultaneously supports Wardle’s critique of siloed research approaches and Gowder’s argument for recognizing user agency in platform governance models.

The Mario Dandy cluster (26.93%), which focused on contextual analysis rather than direct victim support, demonstrates how competing narratives can coexist within the same platform ecosystem without complete polarization. This reflects both what Wardle [28] calls the “evolution of narrative frameworks” and what Gowder [34] terms “competing platform identities” that nonetheless maintain dialogue rather than complete separation.

Together, these patterns suggest that Indonesian digital discourse exhibits sophisticated narrative development and identity formation processes that transcend platform design limitations. Users actively construct meaning systems that connect individual incidents to broader social concerns, creating what Wardle describes as “connective narratives” and what Gowder identifies as “culturally specific governance norms” that emerge from user communities rather than platform policies alone.

4.4. Inter-Group Interaction Characteristics

The analysis reveals sophisticated patterns of inter-group interaction that challenge and extend current theoretical understandings of digital communication. The network’s high resilience (85% path reliability) and balanced distribution of ties (25% strong, 45% medium, 30% weak) suggest a more complex structure than predicted by existing network theories.

This pattern of connection distribution particularly challenges Granovetter’s [38] classic “strength of weak ties” theory. While the theory predicts that weak ties are crucial for information diffusion, our data show a more nuanced reality in digital networks:

• Strong ties (25%) maintain cluster cohesion while allowing external influence;
• Medium ties (45%) facilitate sustained cross-cluster dialogue;
• Weak ties (30%) enable information exposure across ideological boundaries.

The role of bridge nodes (10–15% of accounts) proves particularly significant in challenging platform society theory. While van Dijck et al. [14] emphasize algorithmic governance, our findings suggest that human actors in bridge positions can significantly influence information flows:

• Structural Role
  • Facilitate cross-cluster information flow;
  • Maintain network cohesion;
  • Enable narrative translation between clusters;
  • Support sustained inter-group dialogue.
• Content Function
  • Validate information across cluster boundaries;
  • Contextualize narratives for different audiences;
  • Mediate between competing interpretations;
  • Foster constructive cross-cluster dialogue.

This finding particularly resonates with but also extends Monge and Contractor’s [29] framework. While their theory emphasizes structural aspects of networks, our data reveal the crucial role of human agency in shaping network dynamics:

• Bridge nodes actively curate and translate content;
• They maintain multiple group affiliations while retaining credibility;
• They facilitate dialogue across ideological boundaries;
• They help sustain long-term network resilience.

The high network resilience (85% path reliability) also challenges assumptions about digital network fragility. Contemporary theories often emphasize the volatility of digital networks, yet our data show the following:

• Structural Stability
  • Multiple redundant pathways between clusters;
  • Sustained cross-cluster dialogue;
  • Robust information validation mechanisms;
  • Stable but permeable cluster boundaries.
• Content Sustainability
  • Evolution from reactive to reflective discourse;
  • Development of shared interpretative frameworks;
  • Integration of multiple narrative perspectives;
  • Sustained engagement with complex issues.

4.5. Theoretical Implications

The findings both extend and challenge several theoretical frameworks in digital communication studies, revealing complexities that existing theories do not fully capture. Our analysis suggests a need for more nuanced theoretical models that can account for the unique characteristics of digital discourse in contexts of collectivist cultural dimensions.

Network Society Theory Reconsidered

While’ Castells [8] Network Society Theory provides valuable insights, our findings suggest important modifications:

• Power Dynamics
  • Traditional hierarchies persist within network structures;
  • Institutional authority adapts rather than dissolves;
  • Power flows through multiple, overlapping channels;
  • Cultural factors significantly influence network formation.
• Network Architecture
  • More resilient than predicted (85% path reliability);
  • Higher cross-cluster interaction (20%);
  • Sustained rather than temporary connections;
  • Complex integration of formal and informal networks.

Our findings support Stray et al.’s [15] assertion that managing digital polarization requires understanding how platform mechanics interact with user behavior patterns. The high level of cross-cluster interaction (20%) suggests that platform design can either facilitate or hinder meaningful dialogue across ideological boundaries.

Platform Society Framework Extensions

van Dijck et al.’s [14] platform society concept requires significant modification when applied to Indonesian contexts:

• Algorithmic Governance
  • Platform mechanics influence but do not determine interactions;
  • User agency remains significant in shaping information flows;
  • Cultural practices modify algorithmic effects;
  • Traditional media maintain substantial influence.
• Cultural Adaptation
  • Local communication patterns persist;
  • Platform features adapt to cultural norms;
  • Hybrid forms of authority emerge;
  • Traditional institutions maintain relevance.

Information Disorder Framework Expansion

Tomassi et al. [26] and Wardle and Derakhshan [27] framework requires an extension to account for the following:

• Cultural Context
  • Information verification through cultural networks;
  • Local interpretation of global narratives;
  • Community-based fact-checking mechanisms;
  • Cultural influence on information credibility.
• Network Effects
  • Cross-cluster validation processes;
  • Bridge node verification roles;
  • Community-based truth arbitration;
  • Cultural factors in information assessment.

The findings particularly challenge individualist cultural dimensions’ assumptions about the following:

• Echo Chambers
  • Higher cross-cluster interaction than predicted;
  • More permeable ideological boundaries;
  • Active bridge node facilitation;
  • Sustained cross-ideological dialogue.
• Digital Polarization
  • More complex than binary opposition;
  • Dynamic rather than static clustering;
  • Active inter-group dialogue;
  • Cultural factors moderating polarization.

Our findings extend multiple theoretical frameworks related to digital discourse, platform governance, and information flows. Wardle’s [28] critique of siloed research approaches, Gosztonyi’s [35] analysis of content moderation systems, and Gowder’s [34] concept of democratic platform governance collectively provide a comprehensive lens through which to interpret our results.

The identification of distinct clusters with significant cross-boundary interactions (20% of total activities) challenges core assumptions in these frameworks. While Wardle expresses concern about digital echo chambers and Gosztonyi analyzes how algorithmic content sorting creates information silos, our findings reveal more permeable boundaries between discourse communities in the Indonesian context. This aligns with Gowder’s (2023) [34] emphasis on how platform governance must account for diverse cultural contexts rather than imposing universalized standards.

The Political Buzzer cluster (6.05% of users) exemplifies the intersection of these theoretical perspectives. This cluster represents what Wardle [28] describes as opportunistic engagement that exploits moments of tension, while illustrating what Gosztonyi terms “coordinated inauthentic behavior”, which moderation systems struggle to address. From Gowder’s perspective, this cluster demonstrates what he calls a “democratic deficit” in platform governance, where certain actors can take advantage of algorithmic systems without adequate accountability mechanisms.

The News Media cluster’s role (14.14% of accounts) reflects what Gowder terms “participatory platform identity”—institutional actors adapting to digital environments while maintaining their traditional social functions. Their balanced distribution of interactions across clusters demonstrates what Gowder describes as “governance entities at multiple scales”, mediating between official narratives and community interpretations in what Gosztonyi identifies as “hybrid media systems”.

The temporal evolution of discourse surrounding the Mario Dandy case highlights limitations in current platform governance approaches. The shift from factual reporting to systemic critique demonstrates both the narrative accumulation process that Wardle describes and the ex-post moderation limitations identified by Gosztonyi [35]. This evolutionary process illustrates Gowder’s argument about how platforms struggle with “self-control” and consistent rule enforcement across diverse cultural contexts.

Our cluster analysis reveals how Indonesian digital discourse operates within what Gowder refers to as “cross-cultural conflict management” on platforms. The Pro David (40.12%) and Mario Dandy (26.93%) clusters represent competing narrative frameworks that platforms must moderate without adequate cultural context. This supports Gowder’s assertion that democratic platform governance necessitates cultural sensitivity and local participation in establishing moderation standards.

The bridge nodes (10–15% of accounts) that facilitate cross-cluster dialogue exemplify what Gowder [34] calls “empowered platform users” who transcend algorithmic sorting to maintain community cohesion. Their role challenges the deterministic notion of algorithms as creating inevitable information silos, suggesting that user agency remains significant despite platform governance mechanisms.

4.6. Limitations and Future Research

This study’s limitations must be considered within both theoretical and methodological contexts, while simultaneously pointing toward important future research directions. The platform-specific nature of our analysis, focused exclusively on X (formerly Twitter), presents significant constraints in understanding broader digital communication patterns. While platform society theory provided valuable insights, its application was necessarily limited to X’s specific mechanics and features. This limitation affects not only our theoretical understanding but also the generalizability of our findings to other social media platforms, suggesting a crucial need for cross-platform comparative studies.

Temporal considerations present another significant limitation. Although our study captured immediate and medium-term effects of digital discourse evolution, the bounded time frame potentially missed important long-term patterns and developments. This temporal constraint affects our ability to fully test network evolution theories and understand the sustainability of observed patterns. Future research should address this through longitudinal studies, which can track network evolution, cluster stability, and opinion formation patterns over extended periods.

The methodological focus on social network analysis, while providing robust structural insights, may have missed important qualitative aspects of digital discourse. Our keyword-based sampling approach, while systematic, potentially introduced biases in data collection and analysis. These methodological constraints particularly affect our ability to capture nuanced cultural contexts and meanings. Future research would benefit from mixed-methods approaches that integrate quantitative network analysis with qualitative ethnographic methods to provide a richer, more contextualized understanding of digital discourse dynamics.

Cultural context presents perhaps the most significant consideration for future research. Our application of theoretical frameworks based predominantly on individualist cultural dimensions to Indonesian digital discourse, while yielding valuable insights, also highlights the need for more culturally specific theoretical models. The unique characteristics of Indonesian digital communication patterns, including local interpretations of global platform features and cultural influences on information flow, suggest the need for theoretical frameworks that better account for cultural specificity. This points toward future research opportunities in developing hybrid analytical frameworks that can better capture local communication patterns while maintaining comparative analytical power.

These limitations and future directions suggest a rich research agenda focused on developing more comprehensive, culturally sensitive approaches to understanding digital discourse. Future studies should particularly focus on cross-platform dynamics, long-term pattern evolution, mixed-methods approaches, and culturally specific theoretical development. Such research would not only address the current limitations but also advance our understanding of how digital platforms shape public discourse in diverse cultural contexts.

5. Conclusions

This study provides a comprehensive analysis of how digital networks shape public discourse around youth violence on Indonesian social media, offering significant insights into the complex dynamics of online communication and opinion formation. Through a detailed examination of 264,155 activities from 83,097 accounts, we uncovered sophisticated patterns that challenge traditional assumptions about social media communication and suggest more nuanced understandings of digital public sphere formation.

Our findings reveal a complex network structure characterized by high resilience and significant cross-boundary interaction. The emergence of four distinct clusters, coupled with substantial cross-cluster interaction (20% of total activities) and strong network resilience (85% path reliability), demonstrates that digital discourse exhibits more sophisticated patterns than previously theorized. Particularly noteworthy is the dominance of retweets (68%), indicating a primary focus on information amplification, while the presence of active bridge nodes (10–15% of accounts) plays a crucial role in maintaining network cohesion and facilitating cross-community dialogue.

These findings extend existing theoretical frameworks while highlighting unique characteristics of Indonesian digital discourse, suggesting the need for culturally sensitive approaches to understanding online communication patterns. These insights have significant practical implications for various stakeholders, including digital literacy educators, social media strategists, public communication planners, and youth violence prevention practitioners. The findings suggest that effective intervention in digital spaces requires sophisticated, network-aware approaches that consider both structural dynamics and cultural contexts. Future research should expand upon these findings through cross-platform analysis, longitudinal studies, and integration of qualitative methods to provide a more complete understanding of how digital networks shape public discourse and social responses to youth violence in the contemporary digital age.

This study’s findings highlight the need for more sophisticated approaches to both understanding and governing digital discourse about youth violence. Effective content regulation requires balancing the interests of multiple stakeholders with fundamental rights principles, rather than relying on simplistic content removal strategies. Our analysis suggests that Indonesian digital discourse demonstrates significant resilience and fosters cross-community dialogue, despite potential platform governance constraints.

To build a framework for legitimate content regulation, we recommend the following:

• Develop context-sensitive moderation approaches that address Indonesian cultural and linguistic nuances when evaluating discussions about youth violence.
• Increase transparency regarding how platforms moderate content related to youth violence cases, especially when involving public figures.
• Empower local user communities to participate in platform governance decisions, reflecting Gowder’s call for participatory platform identity and multi-level governance structures.
• Support bridge nodes and legitimate information validators that facilitate cross-community dialogue and verify information.
• Acknowledge the narrative ecosystems shaping digital discourse, rather than focusing solely on individual posts.

These recommendations acknowledge that content regulation will always exist in some form, but establishing legitimate boundaries requires understanding the complex network dynamics revealed in this study.