Modeling Affective Mechanisms in Relaxing Video Games: Sentiment and Topic Analysis of User Reviews

Abstract

The accelerating pace of digital life has intensified psychological strain, increasing the demand for accessible and systematized emotional support tools. Relaxing video games—defined as low-pressure, non-competitive games designed to promote calm and emotional relief—offer immersive environments that facilitate affective engagement and sustained user involvement. This study proposes a computational framework that integrates sentiment analysis and topic modeling to investigate the affective mechanisms and behavioral dynamics associated with relaxing gameplay. We analyzed nearly 60,000 user reviews from the Steam platform in both English and Chinese, employing a hybrid methodology that combines sentiment classification, dual-stage Latent Dirichlet Allocation (LDA), and multi-label mechanism tagging. Emotional relief emerged as the dominant sentiment (62.8%), whereas anxiety was less prevalent (10.4%). Topic modeling revealed key affective dimensions such as pastoral immersion and cozy routine. Regression analysis demonstrated that mechanisms like emotional relief (β = 0.0461, p = 0.001) and escapism (β = 0.1820, p < 0.001) were significant predictors of longer playtime, while Anxiety Expression lost statistical significance (p = 0.124) when contextual controls were added. The findings highlight the potential of relaxing video games as scalable emotional regulation tools and demonstrate how sentiment- and topic-driven modeling can support system-level understanding of affective user behavior. This research contributes to affective computing, digital mental health, and the design of emotionally aware interactive systems.

1. Introduction

In an era characterized by rapid digital acceleration, ubiquitous connectivity, and compressed social rhythms, individuals are increasingly exposed to chronic psychological strain [1]. Contemporary digital lifestyles have contributed to elevated levels of stress, anxiety, and emotional exhaustion. According to the World Health Organization, the prevalence of anxiety disorders increased by 25.6% during the COVID-19 pandemic, reflecting a global surge in emotional instability amid the societal upheaval [2].

Amid these challenges, there is a growing interest in digital practices that promote psychological relief, mindfulness, and a more deliberate pacing of life. Cultural movements such as the Slow Media Manifesto and Digital Detox initiatives underscore a collective search for low-pressure, emotionally restorative digital experiences [3]. Beyond cultural trends, recent advancements in digital health technologies have demonstrated the feasibility of using immersive environments for stress reduction and emotional regulation. For example, psychoeducational interventions delivered through virtual reality (VR) have shown promising effects in reducing anxiety levels and enhancing positive emotional states [4].

Crucially, digital escapism—defined as the psychological retreat from reality through digital media—has been identified as a common coping strategy during times of emotional distress. Studies have shown that video games provide structured spaces for managing stress, boredom, and anxiety, particularly during the COVID-19 pandemic [5,6]. Escapism is not merely avoidance, but may represent an adaptive strategy for restoring emotional balance when considered within the framework of affective game design [7]. In this context, slow-paced games offer a unique form of digital refuge that blends immersion with emotional self-regulation.

Slow-paced digital games have emerged within this broader landscape as promising interventions for emotional health management. Among them, Stardew Valley [8] (available at https://store.steampowered.com/app/413150/Stardew_Valley/ accessed on 10 April 2025)—an independent farming simulation game released in 2016—has been particularly praised for its tranquil design, immersive rural setting, and player-controlled progression. Unlike conventional games that emphasize competition and reflex-based performance, Stardew Valley offers low-stakes, self-paced gameplay that fosters autonomy, relaxation, and emotional support.

While anecdotal reports frequently highlight the game’s “relaxing” and “healing” qualities, large-scale empirical research investigating these claims remains limited. Slow-paced games can be conceptualized as affective digital technologies—systems designed to promote emotional regulation through immersive, user-driven experiences. Related research has shown that emotional computing technologies, including VR-based interventions and adaptive interfaces, can significantly contribute to users’ emotional well-being by facilitating stress mitigation and controlled engagement [9]. Recent frameworks on affective interactivity suggest that emotional game design—through interface cues, pacing, and narrative framing—can significantly shape players’ affective states and engagement [10]. Related research confirms that adaptive and emotionally intelligent game systems improve users’ well-being by fostering emotional awareness and control [11,12].

Grounded in Self-Determination Theory (SDT) [13]—a psychological framework that emphasizes the fulfillment of basic needs for autonomy, competence, and relatedness—this study posits that the mechanics of Stardew Valley fulfill these fundamental needs, providing a digital refuge from the pressures of accelerated modern life.

This study addresses the existing gaps by combining large-scale computational text analysis with psychological theory to investigate the emotional and behavioral impacts of slow-paced gameplay. Using a dataset of nearly 60,000 user-generated reviews in both English and Chinese, we apply sentiment analysis, topic modeling, emotional mechanism tagging, and regression modeling to explore the following research questions:

(1)

To what extent do players express anxiety and emotional relief in their game reviews?

(2)

Are slow-paced game mechanics and pastoral esthetics associated with positive emotional responses?

(3)

Do emotional and mechanical mechanisms correspond to increased playtime, thereby indicating deeper behavioral engagement?

By empirically linking emotional design attributes with user sentiment and behavioral indicators, this study not only contributes to a system-level understanding of affective dynamics in human–digital interaction, but also offers a critical perspective on digital escapism as a coping strategy in slow-paced gaming contexts. In contrast to high-stimulation, fast-paced digital environments, slow games like Stardew Valley provide a form of affective refuge—enabling users to engage in psychological withdrawal from anxiety-inducing realities through immersive, self-paced experiences. The findings offer actionable insights for the development of emotionally supportive interactive systems and underscore the potential of relaxing video games as scalable, low-barrier tools for emotional regulation and everyday escapism within the broader framework of digital mental health and human-centered system design.

2. Literature Review

2.1. Social Acceleration and Anxiety

In contemporary digital society, the theory of social acceleration has emerged as a key framework for understanding emotional exhaustion and growing psychological strain. Rosa defines social acceleration as a structural phenomenon driven by technological advancement, institutional modernization, and the cultural compression of time—all of which displace individuals from sustainable life rhythms [14]. This displacement results in heightened stress, time pressure, and cognitive overload, as individuals are pushed to optimize productivity at the cost of rest, reflection, and relational continuity.

Rosa and Scheuerman (2009) further introduce the concept of desynchronization, which describes how the disintegration of shared temporal structures erodes both interpersonal cohesion and emotional regulation [1]. The constant pursuit of immediacy in digital interactions fragments human attention, undermining the psychological balance between activation and recovery.

Empirical evidence supports this conceptual model. Twenge (2018) documented a marked increase in anxiety and depression among adolescents after 2010, attributing this trend to the widespread use of smartphones and the psychological demands of “always-on” connectivity [15]. In parallel, the World Health Organization reported a 25.6% global rise in anxiety disorders over a single decade, highlighting technological stressors and accelerated digital routines as key contributing factors [2].

These findings reveal a deep structural conflict between emotional well-being and contemporary temporal regimes. As a response, researchers have increasingly turned to digital deceleration as a potential corrective framework. In this context, “slowness” is not merely a cultural critique but a design principle aimed at restoring emotional rhythm and psychological resilience.

2.2. Slow Media and Emotional Design

Slow media theorists advocate for media practices that promote rhythm, reflection, and meaning over speed and efficiency. The Slow Media Manifesto calls for human-centered digital design and intentional interaction, in contrast to algorithmic optimization [3]. While slow values have been explored in analog domains such as slow journalism and slow TV, their relevance to digital games remains largely underexplored. Simulation games like Stardew Valley, characterized by gentle pacing and pastoral themes, offer a promising yet under-theorized case for slow digital design.

McGonigal (2011) argues that well-designed games can cultivate resilience through intrinsic motivation and meaningful engagement, especially when they avoid zero-sum or overly competitive structures [16]. Granic et al. (2014) emphasize that games can facilitate emotion regulation and recovery through flow, mastery, and cognitive absorption [17]. Greenfield (1994) conceptualizes games as cultural technologies that not only entertain but also structure cognitive routines and emotional scripts [18].

Despite these insights, game studies have largely prioritized fast-paced, high-arousal genres. The potential of slow games—defined here as games that support open-ended play, low failure pressure, and immersive repetition—to foster emotional restoration remains empirically under-investigated. This gap presents an opportunity to assess whether slowness itself constitutes an affective affordance in digital interaction [19].

2.3. Casual Gameplay and Behavior

A growing body of literature has begun to explore the affective benefits of casual and non-competitive gameplay. Snodgrass et al. (2011) found that games like FarmVille provide psychological relief by offering players a predictable and low-pressure environment, facilitating what they term “technologies of absorption” [19]. Similarly, Yang et al. (2018) analyzed user behavior in Ant Forest, a gamified environmental app, and concluded that behaviorally repetitive, nature-themed digital tasks promote self-regulation and emotional restoration [20], aligning with broader efforts to understand play as a process of affective feedback [21,22].

Pendergrass Boomer et al. (2023) contributed to this line of inquiry by developing and testing a digital health game designed to prevent anxiety and substance misuse in adolescents. Their findings indicate that emotionally oriented gameplay can produce measurable improvements in mental well-being [23]. Additionally, Kneer et al. (2012) demonstrated that emotional reactions to games vary between perceptions of fun and danger, highlighting subjective distinctions in affective responses to digital play environments [24].

These studies suggest that emotional recovery in digital environments may stem not only from narrative or visual esthetics, but from underlying behavioral structures—such as low-stakes repetition and flexible pacing. Such findings align well with slow game design and lay important groundwork for the empirical analysis of emotional relief mechanisms in player discourse.

2.4. Self-Determination Theory and Digital Needs

SDT is a macro theory of human motivation that posits three innate psychological needs—autonomy, competence, and relatedness—as essential for psychological growth, well-being, and intrinsic motivation [13], providing a robust psychological framework for understanding how digital games can fulfill emotional needs. According to Deci and Ryan (2000) [13], autonomy, competence, and relatedness constitute the three essential ingredients for psychological flourishing. When these needs are satisfied through gameplay, individuals experience enhanced intrinsic motivation, emotional balance, and sustained engagement.

Ryan et al. (2006) applied SDT to video game environments, demonstrating that games that promote player control and personalized progression can enhance immersion and satisfaction [25]. Rigby and Ryan (2006) expanded this analysis, suggesting that low-pressure gameplay mechanisms—such as the absence of failure, self-paced goals, and non-punitive systems—encourage sustainable emotional investment [25]. Tamborini et al. (2011) found that media enjoyment strongly correlates with the satisfaction of SDT-aligned needs [15]. Reinecke (2009) emphasized that open-ended games support emotional recovery by enabling mental disengagement and non-directed play [26].

These findings provide a theoretical bridge to understanding why slow-paced games may serve as affective technologies: their open structure and calming mechanics naturally fulfill the SDT dimensions, though few empirical studies have tested this alignment directly.

2.5. Gameplay Duration and Emotion

Beyond game mechanics, playtime has emerged as a significant behavioral factor in emotional outcomes. Kumar et al. (2024) proposed a theoretical and computational framework for understanding how extended gameplay affects emotional states over time. Their study revealed that players often undergo emotional shifts as they progress through different gameplay phases, with specific patterns linked to the completion of key tasks and ambient environment changes [27]. Wagener et al. (2025) found that certain violent games could reduce stress levels, suggesting that emotional outcomes may also depend on genre-specific mechanisms [6].

Bringula et al. (2020) conducted a survey-based study on player affect and found significant correlations between longer gameplay durations and reported anxiety or cognitive overload, particularly among players with unstructured play habits or low self-regulation [28]. These findings highlight that gameplay duration is not emotionally neutral; it can amplify both positive and negative affect depending on the player context and design constraints.

These insights underscore the importance of integrating playtime data into emotional modeling, especially in computational studies using large-scale review data. Analyzing the intersection between sentiment, game mechanics, and behavioral metrics (e.g., hours played) provides a more nuanced understanding of how games mediate emotional states. Additionally, real-world studies of digital interventions confirm that sustained engagement is contingent on emotional responsiveness and system usability [29].

2.6. Emotion Modeling and Game Design

The modeling of player emotion has become an increasingly central concern in computational game research. Foundational works in affective computing have highlighted the importance of integrating physiological signals and emotional inference in interactive systems [30,31,32]. Sekhavat et al. (2020) explored the use of emotion-aware mechanisms in game design, showing that dynamically adjusting content based on player affect can improve user satisfaction and reduce cognitive overload [33]. Toh and Kirschner (2023) proposed incorporating socio-emotional concepts into learning games to enhance their affective interactivity and pedagogical effectiveness [34]. Melhart et al. (2022) contributed the AGAIN dataset, which contains multimodal annotations of player arousal across nine games, enabling the granular modeling of the emotional response during gameplay [35].

Byl (2015) proposed an affective design framework that integrates emotional dynamics into the core elements of gameplay—interface, narrative, and character design—emphasizing how these components jointly structure players’ affective responses [36].

These contributions point toward the methodological potential of integrating textual data (e.g., user reviews) with behavioral indicators (e.g., playtime) and game features (e.g., pacing, task type). For research on slow games like Stardew Valley, such approaches allow scholars to empirically test whether affective benefits emerge from specific design patterns—thereby grounding the discussion of “slowness” in measurable emotional outcomes. Sentiment strength detection models on social platforms provide a scalable reference for user text emotion mining [37].

2.7. Automated Emotional Valuation Frameworks

As digital environments become increasingly emotionally interactive, there is growing demand for automated systems that can assess, interpret, and adapt to user affect in real time. Emotion modeling has evolved from static, post-hoc analysis to dynamic, AI-driven systems that operate on continuous streams of multimodal input.

Recent studies demonstrate how machine learning and artificial intelligence enable real-time emotional evaluation through facial recognition, sentiment tracking, and physiological data integration, such as EEG-based arousal detection [38] and the DEAP dataset for emotion modeling [39]. Ninaus et al. (2019) proposed a facial emotion detection system to enhance affective engagement in game-based learning [12], while Benlamine et al. (2021) developed a rule-based adaptation interface for VR environments based on affective states [11]. These frameworks represent a shift toward adaptive, emotionally intelligent games that are capable of adjusting the difficulty, narrative, or sensory input to match user needs.

Croissant et al. (2023) outline a design-centric framework for affective interactivity in games, emphasizing how emotional design elements—from UI feedback to pacing and ambiance—shape engagement [10]. In doing so, such systems operationalize emotion as both a design variable and a measurement target. Yannakakis and Melhart (2023) provide a comprehensive review of affective game computing, introducing a four-stage “affective loop” (perception, detection, modeling, and adaptation) as a foundational model for automated emotional assessment in games [32].

These methods underscore a methodological opportunity to integrate user-generated text (e.g., reviews), behavioral data (e.g., playtime), and emotion-aware design features into comprehensive frameworks of emotional value. Laaber et al. (2024) showed that digital maturity correlates with well-being through the satisfaction or frustration of basic psychological needs, offering implications for emotion-sensitive game systems [40]. Martínez-Maldonado (2025) explored how gameplay influences decision-making skills, providing further support for the behavioral impact of affective gaming systems [41].

As video games increasingly serve as affective infrastructures, automated emotional valuation becomes essential for sustainable and responsive system design. Review forums themselves, as genre-specific discursive systems, also reflect dynamic emotional exchanges and can be systematically modeled [42]. The AGAIN dataset further enables fine-tuned modeling of emotional arousal in gameplay [35].

2.8. Research Gap

Although research has increasingly connected digital games to emotional health, the affective impact of slow-paced play remains empirically underexplored. Most existing studies rely on qualitative interviews or limited samples [19], without leveraging the scale and nuance of player-generated data. Basari et al. (2013) [43] and Wang et al. (2023) [44] show that opinion mining and sentiment analysis can effectively identify emotional themes in large text corpora. Advances in topic modeling frameworks, such as probabilistic LDA [45] and neural BERTopic [46], provide the necessary computational tools for such analysis. These methods provide scalable techniques for extracting thematic structures from large-scale textual data [45,47]. Karlsen and Elgesem (2010) [48] advocate for “computational play studies,” where player reviews are treated as valuable behavioral and emotional data. This approach is grounded in well-established sentiment analysis techniques, drawing on foundational work in opinion mining and affective text classification [49,50].

In response to the increasing emphasis on methodological rigor and conceptual clarity, this literature review integrates both the foundational definitions and empirical applications of the key techniques such as sentiment analysis, topic modeling, and affective system design—thereby addressing the existing descriptive gaps and ensuring a robust analytical foundation for the present study.

To address this gap, the present study analyzes nearly 60,000 Stardew Valley reviews using a mixed-methods framework comprising sentiment analysis, dual-stage topic modeling, mechanism tagging, and regression modeling. This approach aims to empirically demonstrate how slow game design functions as an affective technology and supports digital emotional sustainability.

This study builds on established practices in text mining and behavioral modeling, drawing from machine learning foundations [51] and affect-aware system design [32].

3. Materials and Methods

3.1. Research Framework

To ensure coherence between the research design and analytical procedures, this study formulated the following research questions (RQs):

• RQ1: To what extent do players express anxiety and emotional relief in their game reviews?
• RQ2: Are slow-paced game mechanics and pastoral esthetics associated with positive emotional responses?
• RQ3: Do emotional and mechanical mechanisms correspond to increased playtime, indicating deeper behavioral engagement?

To address these questions, we developed a four-module computational framework, with each module aligned with specific RQs:

• Dual-Stage Topic Modeling (Module 1): Identifies dominant themes and emotional vocabulary across the corpus to provide context for RQ1 and RQ2. Latent Dirichlet Allocation (LDA) was applied to both the full corpus and a relaxation-related subset to extract latent semantic themes. Although models such as NMF, Top2Vec, BERTopic, and CorEx offer complementary strengths in interpretability, semantic depth, or temporal modeling, LDA was selected due to its probabilistic structure, well-established performance, and transparency in generating interpretable topic-word and document-topic distributions [45,52]. This makes LDA particularly suitable for thematic comparison across large corpora.
• Sentiment Classification (Module 2): Quantifies expressions of anxiety, relief, and neutrality, directly addressing RQ1. We used a GPT-based large language model to classify each review into one of three emotional states—anxiety, relief, or neutral—based on zero-shot contextual understanding. This method outperforms traditional lexicon-based and supervised models in interpreting informal, multilingual, and affectively nuanced expressions, especially without requiring domain-specific labeled training data [52,53].
• Mechanism Tagging (Module 3): Connects player sentiment to gameplay structures, particularly slow-paced mechanics and pastoral esthetics, addressing RQ2. Using the same model architecture, each review was annotated with one or more gameplay-related mechanism tags (e.g., emotional relief, escapism, slow-paced design). The zero-shot, multi-label classification capability of GPT is well-suited for identifying latent experiential patterns embedded in user discourse.
• Regression Analysis (Module 4): Tests whether emotional or mechanical features predict playtime, serving as a behavioral indicator for RQ3. We implemented multiple linear regression models to explore how emotional states and gameplay-related mechanisms may be associated with playtime. Rather than focusing solely on statistical significance, our analysis aimed to identify potential patterns linking emotional experiences and mechanical features to behavioral engagement. This allowed us to evaluate whether specific emotional or mechanical experiences tend to correspond with a longer playtime, thereby offering empirical insight into RQ3.

This study analyzed a dataset of approximately 60,000 user reviews of Stardew Valley collected from Steam [8] (https://store.steampowered.com/app/413150/Stardew_Valley/ accessed on 10 April 2025), the leading global platform for digital PC games. Reviews were sourced in both English and Chinese and translated into English using a neural machine translation model. Metadata for each review included language, playtime (in hours), and recommendation status. We excluded reviews shorter than 20 characters or identified as spam, off-topic, or bot-generated.

While future research may explore more advanced methods—such as Dynamic Topic Modeling for temporal analysis [54], embedding-based models like BERTopic for semantic precision [46], or hybrid clustering to enhance coherence [43]—our chosen framework emphasizes methodological scalability, interpretability, and alignment with behavioral outcomes, which suits the exploratory and empirical goals of the present study.

The analytical process is illustrated in Figure 1, which provides an overview of the computational framework employed in this study.

3.2. Data Collection and Platform Background

Steam [55] is the largest digital distribution platform for PC games. According to IHS Screen Digest, it accounted for approximately 75% of the global market as early as 2013 [56]. Although more competitors have emerged since then, Steam remained a dominant hub for global PC gaming through the 2020s. By 2021, the platform hosted over 34,000 game titles and had more than 132 million monthly active users [56]. Its verified-purchase review system allows players to post textual feedback linked with behavioral metadata such as cumulative playtime, language, and recommendation status (e.g., “Recommended”). In addition, reviews support community-driven emotional tagging, including community votes such as “Helpful” and “Funny,” thereby enriching the platform’s potential for large-scale sentiment and engagement analysis.

This study utilized publicly available user review data from the Steam platform [8]. All data were anonymized and did not involve any personal identifiable information, ensuring compliance with ethical research standards. The data collection process adhered strictly to the terms of use of the Steam API, and no privacy violations were involved.

A total of 63,818 raw reviews were collected for Stardew Valley using a Python-based crawler and the Steam API [8]. After translation and preprocessing, 57,911 valid entries were retained for analysis, reflecting a 9.3% reduction due to the removal of duplicates, irrelevant content, or short comments (<20 characters).

The preprocessing phase involved a series of systematic steps to ensure that the textual data was clean and analytically tractable. First, all user reviews were converted to lowercase, and non-alphanumeric symbols were removed to standardize the input. Next, verb forms were lemmatized to their base forms, while plural nouns were converted to singular, enhancing the consistency across variations in word usage. Following this, the text was tokenized into individual words, and common stopwords were filtered out to reduce noise. Finally, reviews were filtered using a predefined whitelist of game-specific keywords, which ensured that the retained dataset aligned with the thematic focus of the study.

To gain a preliminary understanding of the dominant lexical patterns in the corpus, word frequency analysis was conducted after preprocessing. A word cloud was then generated to visualize the most salient terms used in user reviews. As illustrated in Figure 2, frequently occurring words such as game, play, farm, relax, cozy, and love highlight the emotionally soothing and pastoral dimensions of the Stardew Valley gameplay experience.

3.3. Topic Modeling (LDA)—Full Corpus

This study employed Latent Dirichlet Allocation (LDA), an unsupervised machine learning technique, to analyze and extract latent thematic patterns within user reviews of the game. LDA is particularly well-suited for large-scale text analysis, as it models each document as a probabilistic mixture of topics, and each topic as a distribution over words [57].

The LDA model was implemented using the Python Gensim library, applied to a pre-processed corpus of 57,911 reviews. The core LDA model assumes that each document (user review) is generated by a mixture of topics, where each topic has a set of keywords with different probabilities. The general formula for this probabilistic model is $P\left(w_j|D_i\right)={\sum }_{k=1}^{K}P\left(w_j|T_k\right)P\left(T_k|D_i\right)$, where $w_j$ represents the words in the corpus, $D_i$ represents the individual documents, and $T_k$ represents the topics associated with the documents, forming a three-layer probabilistic structure. By performing probability calculations for the distribution of words, topics, and documents, the topic distribution of the entire text corpus can be inferred.

The number of topics was optimized through iterative testing, balancing coherence and perplexity scores. As illustrated in Figure 3 and Figure 4, evaluation metrics were plotted against the number of topics to identify the optimal model configuration. These visualizations reveal a turning point at seven (7) topics, where coherence is maximized without a substantial increase in perplexity. This configuration ensured a meaningful division of semantic themes while maintaining manageable model complexity. The detailed interpretation of the extracted topics, along with their associated sentiment patterns and affective implications, will be presented in Section 4.1.

The Intertopic Distance Map (Figure 5) provides a global view of how topics are distributed in the semantic space. Each topic is represented as a circle, with size reflecting its prevalence and distance indicating semantic dissimilarity. This visualization helps assess the distinctiveness and potential overlap between topics, thereby supporting the evaluation of thematic coherence across the corpus.

3.4. Topic Modeling (LDA)—Slow-Paced Subset

To extract a relaxation-specific subset from the corpus, we defined a list of 18 slow-paced keywords, including the following: relax, relaxing, relaxed, chill, calm, peaceful, slow, no rush, no pressure, no time limit, unwind, cozy, soothing, stress-free, immersive, gentle, quiet, and tranquil.

A binary label (target) was assigned to each review using a simple keyword-matching function in Python (version 3.9.7). Reviews containing at least one slow-paced keyword were marked as 1; all others were marked as 0. Out of 57,911 total reviews, 6802 entries (≈11.7%) were classified as “slow-paced related” and used to construct the subcorpus.

We also applied LDA to the slow-paced subset using the Gensim library. The optimal number of topics was determined through the systematic evaluation of perplexity and coherence scores. As illustrated in Figure 6 and Figure 7, we observed a balance between semantic interpretability and model simplicity when the number of topics was set to four (4). This configuration offered optimal coherence while maintaining clarity in topic distinctions, supporting the interpretability of relaxation-related themes.

For the slow-paced subset, topic visualization was also conducted using pyLDAvis, as shown in Figure 8. This facilitated a closer examination of topic distributions and salient terms related to relaxation-oriented themes. In particular, the Top-30 Most Salient Terms plot (Figure 8) identifies the terms that contribute most to topic interpretability by integrating both frequency and relevance. This visualization highlights the key lexical features that players use to describe their emotional responses and gameplay experiences, offering valuable insight into the affective dynamics of slow-paced digital environments.

3.5. Sentiment Analysis

To examine the emotional landscape of user experiences with relaxing video games, we employed large-scale sentiment classification on all 57,911 preprocessed reviews. This step was designed to address RQ1: Do player sentiments reflect anxiety and relief?

3.5.1. Model Choice and Motivation

Rather than relying on traditional lexicon-based approaches, we used OpenAI’s large language model (LLM), specifically gpt-4o-mini, to ensure higher semantic accuracy and contextual understanding. This method captures latent emotional cues, sarcasm, and subtle valence shifts often missed by standard classifiers. Given the contextual richness of review texts, GPT-based annotation offers flexible and high-accuracy labeling, though it is subject to interpretive variation.

3.5.2. Prompt Engineering and Classification Logic

To analyze the emotional impact of the reviews, each review was processed using a structured prompt that directed the model to categorize it into one of three emotional states: Anxiety Expression, emotional relief, or neutral. Anxiety Expression encompassed references to stress, burnout, emotional overwhelm, or mental strain, capturing negative emotional experiences. Emotional relief included expressions of relaxation, enjoyment, emotional healing, or psychological relief, reflecting positive emotional outcomes. Neutral covered factual, evaluative, or gameplay-related content lacking a clear emotional tone. The model was prompted with the instruction: “Please classify the following content into one of three emotional categories: Anxiety Expression, Emotional Relief, or Neutral. Provide two fields: (1) Sentiment State and (2) Explanation.” For example, a sample output might read: Sentiment: Emotional relief; Explanation: The comment reflects enjoyment and relaxation, suggesting a positive emotional state. This structured approach ensured consistent and interpretable classification of the emotional content within the reviews.

3.5.3. Batch Processing and Storage

A Python-based automation pipeline was implemented using the OpenAI Application Programming Interface (API). For each review, the API-generated output was saved locally as a .txt file, with two fields extracted: Sentiment State and Explanation. Sentiment labels were normalized into lowercase categories: anxiety, relief, and neutral.

The analysis of 57,911 reviews revealed a significant distribution of emotional sentiments, with 36,371 reviews (62.8%) classified as emotional relief, reflecting expressions of relaxation, enjoyment, or emotional healing. In contrast, 15,518 reviews (26.8%) were categorized as neutral, encompassing factual or gameplay-related content without strong emotional undertones. Only 6022 reviews (10.4%) fell under Anxiety Expression, indicating references to stress, burnout, or mental strain. This distribution underscores a clear predominance of relief-oriented sentiments, supporting the hypothesis that slow-paced game design, as exemplified by Stardew Valley, fosters relaxation and emotional well-being among players.

3.6. Mechanism Tagging

To address RQ2 (Are slow-paced mechanisms or rural settings tied to positive emotions?), this section employed mechanism-level tagging to annotate the emotional and experiential features in each review. We utilized an LLM to infer the psychological mechanisms underlying user feedback.

3.6.1. Motivation and Conceptual Framework

Building on SDT and previous studies on digital well-being, we developed a tagging framework encompassing six theoretically grounded gameplay mechanisms. Each mechanism represents a potential pathway through which gameplay might alleviate anxiety or enhance psychological well-being. These mechanisms are defined in detail in

Table 1. Definition of mechanism tags.

Label	Definition
Anxiety Expression	Expressions of stress, tiredness, emotional exhaustion, or negative states
Emotional Relief	Descriptions of calm, joy, comfort, or emotional healing
Slow-paced Gameplay	References to relaxed progression, absence of pressure, or low-speed design
Pastoral Healing	Rural symbols such as farming, nature, animals, or a simple life
Control/Low Stakes	Elements of autonomy, sandbox systems, or minimal failure consequences
Escapism	Desires to forget reality or immerse oneself in alternative worlds

below.

3.6.2. Implementation via GPT

To annotate the dataset with gameplay-related mechanisms, we employed OpenAI’s gpt-4o-mini model, utilizing a structured prompt to guide a multi-label classification process. The prompt was carefully designed to enable the model to assign one or more relevant mechanism tags to each user review, capturing the diverse gameplay experiences expressed in the text. Additionally, the model was instructed to provide a brief textual justification for each assigned tag, ensuring transparency and interpretability in the annotation process. The prompt was formulated as follows: “Please classify the review into one or more of the following tags (multi-label allowed): Anxiety Expression, emotional relief, slow-paced gameplay, pastoral healing, Control/Low Stakes, escapism. Return both the tags and a brief explanation.” For instance, a sample output from the model might read: Tags: slow-paced gameplay, pastoral healing, escapism; Explanation: The comment reflects enjoyment of a farming game with slow mechanics, suggesting immersion and escape from real-life stress. This annotation strategy enabled the systematic identification of gameplay mechanisms—such as slow-paced design and pastoral healing—that support emotional relief and sustained engagement, laying a solid foundation for evaluating the therapeutic potential of relaxing video games.

3.6.3. Pipeline and Storage

To process the 57,911 reviews, a Python-based pipeline was developed to systematically iterate over the dataset, ensuring the efficient and reliable handling of the data. Each review was processed using the LLM, and the corresponding output was saved as a local text file to facilitate reproducibility and enable error recovery. From the LLM output, two key components were extracted: mechanism tags, represented as a string of labels, and Explanation, providing a natural language justification for the assigned tags. To prepare the data for analysis, the label string was binarized into six distinct Boolean columns, each corresponding to one of the mechanism tags (Anxiety Expression, emotional relief, slow-paced gameplay, pastoral healing, Control/Low Stakes, escapism), with a value of 1 indicating the presence of a tag and 0 indicating its absence. This multi-label assignment approach allowed each review to be associated with multiple mechanisms simultaneously, capturing the complexity of player experiences and enabling a comprehensive analysis of the gameplay elements contributing to emotional outcomes in relaxing video games.

3.6.4. Results Summary and Visualization

Out of all the reviews, 55,599 (95.9%) were successfully assigned one or more mechanism tags. Figure 9 displays the distribution of tag frequencies.

3.7. Regression Modeling

To investigate the influence of slow-paced games, such as Stardew Valley, on the total playtime (the dependent variable), we conducted three sets of multiple linear regression analyses. The primary goal was to assess whether emotional expression and gameplay mechanisms predict player engagement, as measured by the total playtime.

Prior to model estimation, we performed a series of statistical diagnostics to ensure the validity of our regression approach. Shapiro–Wilk tests revealed that categorical predictors such as topic category and emotion state significantly deviated from normality (p < 0.05), and Levene’s test indicated unequal variances across groups (p < 0.001). Consequently, Kruskal–Wallis and Mann–Whitney U tests were used for group comparisons. The results confirmed that both topic categories and emotional states were significantly associated with the total playtime (p < 0.001), with pairwise comparisons (Bonferroni-corrected) showing meaningful differences across most groups. Binary predictors, including language, emotional tags, and gameplay mechanisms, were also assessed using Mann–Whitney U tests, most of which yielded significant differences in the median playtime between groups (p < 0.05). For continuous predictors, Spearman’s correlation tests indicated statistically significant but modest associations with playtime (ρ < 0.2, p < 0.001). These findings justified the inclusion of all the predictor types in the regression models.

After that, three distinct models were constructed, each incorporating different combinations of independent variables to systematically evaluate their effects on playtime:

• Model 1 (Full Specification) included all the relevant predictors to provide a comprehensive analysis. These predictors comprised player characteristics (e.g., number of games owned, number of reviews), gameplay mechanisms (e.g., emotional relief, pastoral healing, escapism, Control/Low Stakes), sentiment categories (Sentiment_anxiety, Sentiment_relief), and topic distributions (Topic 1–7). Additionally, control variables such as the number of reviews, language, review length, and recommendation status were included. This holistic model aimed to elucidate the combined influence of player attributes, gameplay mechanics, emotional sentiment, and contextual factors on player engagement, as reflected by the total playtime.
• Model 2 (Mechanism-Only) focused exclusively on the six gameplay mechanisms and two sentiment labels, isolating their independent effects on player engagement. The predictors in this model included mechanisms such as emotional relief, pastoral healing, escapism, and Control/Low Stakes, alongside sentiment labels like Sentiment_anxiety and Sentiment_relief. By excluding control variables and topic distributions, Model 2 sought to evaluate the direct relationship between gameplay mechanisms, emotional expression, and playtime, free from the influence of extraneous factors.
• Model 3 (Mechanism + Sentiment + Control Variables) extended Model 2 by incorporating control variables, such as language, review length, and positive rating, in addition to the gameplay mechanisms and sentiment labels. This model was designed to examine the extent to which these contextual factors, when considered alongside gameplay mechanisms and emotional sentiment, could further account for the variance in playtime, thereby providing a more robust understanding of player engagement.

All the models were estimated using Ordinary Least Squares (OLS) regression. Residual diagnostics confirmed approximate normality (Shapiro–Wilk p > 0.05), independence (Durbin–Watson ≈ 1.95), and acceptable multicollinearity (VIF < 10 for all predictors). Heteroskedasticity, where present, was addressed using robust standard errors. Figure 10 presents the Q–Q plot of standardized residuals from Model 1, illustrating their alignment with the normality line and supporting the validity of the normality assumption. We also implemented LASSO regression for variable selection and verified that the key predictors remained significant in the reduced model, thereby ensuring model parsimony and interpretability.

4. Results and Analysis

4.1. Topic Modeling Results

4.1.1. Model of Full Corpus

To explore the dynamics of emotional engagement in Stardew Valley, we applied Latent Dirichlet Allocation (LDA) to a full corpus of nearly 60,000 user-generated reviews, collected from the Steam platform. The model was implemented using Python’s Gensim library, and the results were visualized using pyLDAvis for enhanced topic exploration.

After iterative testing and coherence score evaluations, we determined that the optimal number of topics for the model was seven (7). This configuration offered a balance between model complexity and semantic clarity, ensuring that the topics were both distinguishable and meaningful. To further validate this choice, we evaluated perplexity scores, which confirmed the robustness of the seven-topic solution.

Each topic was identified based on its top 30 keywords, which were then used to assign semantic labels reflecting the underlying themes of user discourse. These thematic interpretations are summarized in

Table 2. LDA topics from full corpus and their semantic interpretation.

Topic	Top 10 Keywords
Topic 1: Emotional Release and Relaxation	game, great, relax, cozy, update, haley, perfect, work, free, masterpiece
Topic 2: Social and Emotional Connection	love, farm, concernedape, farmer, fantastic, comfort, absolute, peak, playful, couple
Topic 3: Game Mechanics and Player Evaluation	game, good, play, hours, farming, buy, amazing, mods, update, years
Topic 4: Game Experience and Entertainment	fun, play, friend, super, recommend, addictive, modding, lot, highly, hours
Topic 5: Game Expansion and Player Feedback	game, good, play, hours, farming, buy, amazing, mods, update, years
Topic 6: Rural Life and Game Experience	valley, Stardew, play, life, day, sim, star, real, labor, live
Topic 7: Game Experience and Player Feedback	game, farm, time, feel, enjoy, character, story, gameplay, experience, recommend

.

These topics reflect the underlying emotional mechanisms, particularly focusing on aspects such as relaxation, social interaction, and emotional engagement—key themes associated with emotional relief and healing as highlighted in the introduction. Notably, themes centered on “Emotional Release and Relaxation” and “Social and Emotional Connection” were consistently linked with keywords typically associated with slow-paced gameplay mechanics, such as “relax,” “cozy,” and “stress-free.” These findings directly correspond to our first research question, suggesting that players indeed express emotional relief and relaxation in their reviews. Additionally, topics such as “Game Expansion and Player Feedback” and “Game Experience and Entertainment” further highlight how players find satisfaction and enjoyment through immersive gameplay, social interactions, and the overall game experience, which reinforces the emotional healing aspects of the game.

The analysis of the topic distribution in Figure 11 provides valuable insights into how players engage with different aspects of gameplay and emotional experience. The distribution highlights a clear hierarchy of emotional and gameplay-related themes, reflecting the varying levels of interest and emotional investment from players.

Topic 3 emerged as the most prominent, with 12,389 reviews associated with it, suggesting that themes related to emotional relief and healing strongly resonate with players. Topic 7, related to game experience and player feedback, also had a high frequency with 11,576 reviews, indicating that players are highly engaged with their overall experience and the immersive aspects of the game. Topic 5, focused on game mechanics and player evaluation, had 9349 reviews, highlighting significant interest in the gameplay and player-generated feedback. In contrast, topics related to more niche aspects, such as Topic 4 (with 3577 reviews), were less prevalent, suggesting that themes like game entertainment and addiction appeal to a smaller subset of players. Overall, the analysis reveals that Stardew Valley effectively addresses players’ emotional needs through its slow-paced mechanics, pastoral esthetics, and social interactions. Themes associated with relaxation, emotional relief, and social connection were most prominent, confirming the significance of these elements in promoting a positive gaming experience. The findings also suggest that while the game’s slow pace fosters emotional relief and a longer playtime, a subset of players focuses on its entertainment and addictive qualities. These results provide a comprehensive understanding of how the game facilitates emotional healing and engagement through its gameplay mechanics.

4.1.2. Model of the Slow-Paced Subset

The LDA model was applied to a subset of 6802 reviews (≈11.7% of the total corpus) that were filtered based on 18 relaxation-related keywords, such as “chill,” “calm,” “no rush,” and “unwind.” The optimal number of topics for this slow-paced subset was determined to be four (4), based on iterative testing and coherence score evaluations. These topics represent key aspects of the player experience related to relaxation, emotional relief, and gameplay immersion.

Based on the top keywords for each topic, we manually labeled the four themes as shown in

Table 3. Semantic interpretation of four LDA topics in the slow-paced review subset.

Topic	Theme Name	Top 10 Keywords
Topic1	Serene Daily Routines	game, relax, play, time, fun, good, day, great, long, lot
Topic2	Cozy and Immersive Ambiance	game, cozy, play, chill, fun, love, relaxing, recommend, great, nice
Topic3	Immersive Farmstead Experience	game, Stardew, valley, farm, experience, gameplay, feel, character, player, community
Topic4	Tranquil Village Connections	life, Stardew, valley, farm, game, feel, time, town, day, love

.

These topics reflect the emotional mechanisms underlying the slow-paced aspects of Stardew Valley, with particular emphasis on relaxation, emotional relief, and social interaction—core components tied to psychological well-being and emotional healing, as highlighted in the introduction. Themes like “Serene Daily Routines” and “Cozy and Immersive Ambiance” were strongly linked with keywords that evoke slow-paced gameplay, such as “relax,” “cozy,” and “chill.” These findings directly answer our first research question, providing compelling evidence that players express emotional relief and relaxation in their reviews.

In addition, themes like “Immersive Farmstead Experience” and “Tranquil Village Connections” highlight how players derive satisfaction and joy from the peaceful, immersive environment of Stardew Valley. Keywords such as “life,” “Stardew,” “valley,” and “farm” underscore the emotional connections which players form with the idyllic, rural setting and the game’s gentle, low-pressure mechanics. This analysis reinforces the idea that Stardew Valley fosters deep emotional engagement in relaxing video games, offering players an emotional refuge from the stresses of daily life.

By analyzing themes focused on emotional release, slow-paced game mechanics, and social interaction, this study confirms that Stardew valley provides a meaningful emotional experience, contributing to player well-being and serving as a therapeutic escape.

Each review in the subset was assigned a dominant topic, and their distribution is shown in Figure 12.

Topic 2 emerged as the most prominent, with 3312 reviews associated with it, indicating that themes related to cozy and immersive ambiance strongly resonate with players. Topic 3, focused on rich farmstead adventures, had 1385 reviews, highlighting that players are highly engaged with the farm-related gameplay aspects, such as planting, exploring, and interacting with characters. Topic 1, centered on serene daily routines, garnered 1335 reviews, emphasizing players’ appreciation for the low-pressure, peaceful pace of the game. Topic 4, focused on tranquil village connections, had 770 reviews, showing that while social connections and the quiet rural environment are important, they appeal to a smaller subset of players.

The analysis reveals that Stardew Valley’s most popular themes are those centered around relaxation, emotional relief, and immersion in a calming atmosphere. The significant proportion of reviews linked to “Cozy and Immersive Ambiance” underscores the game’s primary strength in providing players with an emotionally soothing environment. The lower proportions for “Tranquil Village Connections” suggest that while the social aspects of the game are valued, they are of secondary importance compared to the overall atmosphere and farm-based gameplay. This supports the notion that the game’s slow-paced mechanics, rural esthetics, and social interactions collectively create an emotionally fulfilling experience.

4.1.3. Similarity Analysis of Full Corpus and Slow-Paced Subset Topics

The heatmap presented in Figure 13 visualizes the similarity scores between topics derived from the full corpus (Model A) and the slow-paced subset (Model B). These scores indicate the degree of similarity between pairs of topics, reflecting how closely related their content is. Higher values suggest stronger connections between topics, while lower values indicate weaker associations.

The key findings from Figure 13 are highlighted below:

• Strong Correlation between Certain Topics: There is a notable strong similarity between topics from the full corpus and those from the slow-paced subset. For example, the topic “All comments1” from the full corpus has a high similarity score of 0.71 with “Slow rhythm1” from the slow-paced subset, suggesting that themes related to emotional release and relaxation in the full corpus strongly resonate with slow-paced game elements such as calm gameplay mechanics and immersive experiences.
• Correlations between Game Mechanics and Emotional Relief: The pair, “All comments3” and “Slow rhythm2,” shows a similarity score of 0.82, indicating a strong correlation between game mechanics, player evaluation, and slow-paced features such as peaceful gameplay and daily routines. This reflects how game mechanics designed for emotional relief are a prominent feature across both datasets.
• Weaker Correlation for Niche Topics: Some topics, such as “All comments4” and “All comments5,” show weaker correlations with the slow-paced subset. These topics are more related to entertainment-focused aspects like game addiction or pure entertainment, which appear less connected to the slower-paced, therapeutic features of the game.
• Internal Similarities within Slow-Paced Themes: Within the slow-paced subset, topics like “Slow rhythm3” and “Slow rhythm4” exhibit higher internal similarity, with a score of 0.72. This demonstrates that themes focusing on rural life and emotional engagement are closely associated with other slow-paced game elements, such as community interaction and immersive farming activities.

Overall, this heatmap highlights how slow-paced game mechanisms, particularly those related to emotional release, social interaction, and peaceful gameplay, are deeply intertwined with the broader emotional experiences expressed by players. These findings provide a comprehensive understanding of how slow-paced mechanics contribute to emotional healing and well-being, aligning with the therapeutic and sustainable aspects of gaming.

4.2. Sentiment Analysis Results

4.2.1. Distribution of Sentiment Categories

This section analyzes the sentiment distribution across different topics, focusing on the emotional tendencies expressed by players in their reviews of Stardew Valley. The sentiment categories include anxiety, neutral, and relief, each representing a distinct emotional response to game experience. The final sentiment distributions were visualized using a count-based bar chart (Figure 14). This chart aids in interpreting the emotional prevalence within the player community and provides quantitative support for the mental health design affordances hypothesized.

As illustrated in Figure 14, the majority of reviews (62.8%) expressed emotional relief, highlighting the game’s role as a soothing and enjoyable escape from stress. A notable portion of the reviews (26.8%) remained neutral, offering objective descriptions of the game without significant emotional shifts. Meanwhile, a smaller fraction (10.4%) reported feelings of anxiety, possibly stemming from in-game challenges or stress induced by specific events within the game.

We also present the findings from our sentiment analysis, addressing the core research questions related to player emotions, the role of slow-paced game mechanics and pastoral esthetics, and the relationship between these elements and emotional engagement in relaxing video games.

As shown in Figure 15, the results reveal that most player reviews in Stardew Valley express emotional relief rather than anxiety, aligning with the general therapeutic nature of the game. Emotional relief consistently forms the highest proportion across all topics, with Topic 2 (Cozy and Immersive Ambiance) showing a significant 64.1% of reviews expressing relief. In contrast, anxiety is notably less prevalent, particularly in topics such as Game Expansion and Player Feedback (Topic 5), where only 7.5% of reviews are associated with anxiety.

Slow-paced game mechanics and pastoral esthetics are strongly associated with positive emotional sentiments, particularly relief. Themes like Cozy and Immersive Ambiance (Topic 2) and Rural Life and Game Experience (Topic 6) are intrinsically linked to keywords such as “cozy,” “chill,” and “relaxing,” which directly contribute to the overall positive emotional experiences of players. For instance, Topic 2, which highlights the tranquil, comforting aspects of the game, shows a significant 64.1% of reviews associated with emotional relief, while only 4.2% express anxiety.

These findings directly support the hypothesis that players predominantly use Stardew Valley as a means of emotional relief. The game’s relaxing and calming environment, as reflected in the high relief sentiment across topics, offers a meaningful escape for players, aligning with the findings from previous studies that emphasize the therapeutic potential of slow-paced games. This analysis supports the idea that the game’s slower pace and peaceful rural esthetic are central to fostering a positive emotional environment for players. The soothing mechanics of farming, social interactions, and exploration in a serene setting contribute substantially to emotional relief, confirming the association between these game elements and positive sentiments.

4.2.2. Sentiment Keyword Visualization

In this section, we present the visualization of sentiment-related keywords based on the emotional analysis of the player reviews. The word cloud generated from the reviews categorized under the relief sentiment (Figure 16) provides valuable insights into the words most commonly associated with the emotional relief expressed by players.

The word cloud presented here visualizes the most frequently mentioned keywords associated with the “relief” sentiment in slow-paced reviews of Stardew Valley. Prominent keywords such as “game,” “fun,” “play,” “Stardew,” “farm,” and “love” suggest that players primarily associate emotional relief with the immersive gameplay experience and positive social interactions within the game. Words like “relax,” “cozy,” “chill,” and “peaceful” further highlight the game’s ability to provide a calming, stress-free environment, consistent with its slow-paced mechanics and pastoral esthetics.

Additionally, the frequent appearance of words such as “hours,” “friend,” “farming,” and “time” indicates that players find enjoyment and satisfaction in the game’s routine activities, contributing to their overall sense of accomplishment. The presence of words like “recommend” and “highly” reflects positive emotional outcomes, as players express a desire to share their enjoyable experiences with others.

The prominence of “love” and “friend” underscores the strong emotional bonds which players form with in-game characters, reinforcing the role of social interaction in fostering emotional relief. Overall, this word cloud visually supports the conclusion that Stardew Valley offers an emotionally supportive environment that promotes relaxation and emotional release, primarily through its core gameplay and social elements.

This sentiment keyword visualization provides a clear representation of how the game’s slow-paced, immersive features contribute to positive emotional outcomes, aligning with the principles of psychological well-being emphasized in previous sections.

4.3. Mechanism Tagging Results

In this section, we present the results of the mechanism tagging analysis, which examines the relationships and frequencies of the different emotional mechanisms identified in the reviews. These mechanisms include “Emotional Relief,” “Slow-paced Gameplay,” “Pastoral Healing,” “Escapism,” and “Anxiety Expression.” The analysis aims to understand how these mechanisms interact with one another and how they contribute to the emotional experiences of players in relaxing video games.

4.3.1. Tag Frequency

The heatmap displayed in Figure 17 illustrates the frequency of co-occurrence between the different mechanisms. The darker cells indicate stronger co-occurrence between pairs of mechanisms, while lighter cells represent weaker associations. A key observation from this heatmap is the particularly strong relationship between emotional relief and pastoral healing, with both mechanisms showing a high co-occurrence frequency (in blue cells), reflecting the game’s ability to provide players with both emotional healing and a tranquil, stress-free environment. Additionally, escapism shows significant co-occurrence with emotional relief and pastoral healing, emphasizing the role of the game as an escape from daily life, providing a source of comfort for players. The slow-paced gameplay mechanism also demonstrates notable associations with emotional relief, suggesting that the leisurely pace of the gameplay contributes to a calming emotional experience for players.

Figure 17 also reveals that Anxiety Expression shows a lower frequency of co-occurrence with the other mechanisms, suggesting that while Stardew Valley addresses anxiety in some reviews, it is less commonly discussed in connection with the game’s core emotional relief mechanisms. This reinforces the idea that the game is primarily associated with alleviating stress and providing a calm, immersive experience.

4.3.2. Co-Occurrence Patterns

The network graph presented in Figure 18 further visualizes the relationships between the emotional mechanisms. In this graph, the nodes represent the different mechanisms, and the edges between them illustrate the strength of their associations. The graph clearly shows that emotional relief and pastoral healing are closely connected, with both mechanisms positioned near each other at the center of the network. This highlights the central role that these mechanisms play in providing an overall emotional experience in relaxing video games.

Additionally, the slow-paced gameplay mechanism is also strongly linked with emotional relief and escapism, reinforcing the idea that the game’s relaxed pace and its role as an escape from reality are fundamental in helping players achieve emotional relief. Anxiety Expression appears more isolated in the network, with weaker connections to other mechanisms. This suggests that while anxiety may be present in some reviews, it is neither as dominant nor closely linked with the overall emotional experiences that the game provides.

4.4. Regression Modeling Results

To examine how emotional expression and gameplay mechanisms affect behavioral engagement, measured by the total playtime, we implemented three sets of multiple linear regression models. Prior to regression, all continuous predictors were tested using the Shapiro–Wilk normality test, which confirmed non-normal distribution (p < 0.001). Heteroskedasticity was detected using the Breusch–Pagan test (p < 0.001) and addressed using robust standard errors. Multicollinearity was evaluated via Variance Inflation Factors (VIFs), with all the variables exhibiting acceptable levels (VIF < 10). LASSO regression was further employed for feature selection, and the final models were re-estimated with retained predictors to enhance parsimony.

• Model 1: All Variables. This model included all the independent variables: gameplay mechanisms, sentiment labels, topic distributions, and control variables. It explained 3.6% of the variance in the total playtime (R² = 0.036). Significant predictors included emotional relief (β = 0.0463, p = 0.001), pastoral healing (β = 0.0715, p < 0.001), escapism (β = 0.1345, p < 0.001), and Control/Low Stakes (β = 0.1941, p < 0.001). Anxiety Expression negatively affected playtime (β = −0.0584, p = 0.001). Control variables such as the number of reviews (β = −0.0797, p < 0.001), language (β = −0.2008, p < 0.001), and review length (β = 0.0199, p = 0.001) were also statistically significant. Positive rating (β = 1.2808, p < 0.001) showed the strongest positive effect on playtime.
• Model 2: Mechanism + Sentiment. This reduced model retained only gameplay mechanisms and sentiment labels, isolating their direct effects. It explained 1.5% of the variance (R² = 0.015). Emotional relief (β = 0.0531, p < 0.001) and escapism (β = 0.1938, p < 0.001) remained significant. Anxiety Expression (β = −0.0456, p = 0.010) showed a weaker but still significant negative effect.
• Model 3: Mechanism + Sentiment + Control Variables. In this extended model, language, review length, and positive rating were added to the previous predictors. It explained 2.7% of the variance (R² = 0.027). Emotional relief (β = 0.0461, p = 0.001), escapism (β = 0.1820, p < 0.001), and positive rating (β = 1.3187, p < 0.001) remained strong predictors. However, Anxiety Expression (β = −0.0273, p = 0.124) and slow-paced gameplay (β = 0.0075, p = 0.621) became statistically insignificant after including contextual factors.

Figure 19 visualizes the topic-specific distributions of playtime. Notably, Topic 6 (“Rural Life and Game Experience”) exhibited the highest median playtime, highlighting the role of immersive and affective game features in sustaining engagement. These results confirm that emotionally supportive gameplay mechanisms, particularly escapism and pastoral relief, contribute to a longer playtime and deeper behavioral involvement, even after controlling for sentiment and user characteristics.

5. Discussion

This study set out to explore how slow-paced game mechanics and emotionally supportive design features in Stardew Valley influence players’ emotional responses and behavioral engagement. Through the integration of topic modeling, sentiment analysis, mechanism tagging, and regression analysis, we identified a consistent pattern of emotional relief and increased playtime associated with relaxation-oriented gameplay themes. These findings contribute to a growing body of the literature on affective game design and digital mental health by providing empirical evidence for the therapeutic potential of slow-paced games. In this discussion, we critically interpret the key results in light of the existing research, examine their broader psychological and behavioral implications, and outline the potential directions for future study.

5.1. Emotional Design, Player Sentiment, and Psychological Needs

Topic modeling revealed that players frequently referred to experiences of relaxation, immersion, and emotional relief, suggesting that these themes are central to the perceived emotional value of the game. Dominant patterns such as “Emotional Release and Relaxation” and “Cozy and Immersive Ambiance” emerged consistently across both the full dataset and the slow-paced review subset. These findings indicate that slow-paced mechanics are deeply embedded in the emotional structure of gameplay, rather than being merely esthetic choices.

To interpret these emotional dynamics more systematically, we draw on SDT [13], which posits that autonomy, competence, and relatedness are fundamental psychological needs. In Stardew Valley, autonomy is fostered through player-directed pacing and flexible goal setting; competence develops through agricultural progress and skill advancement; and relatedness emerges through interactions with non-playable characters (NPCs) and the virtual community. These findings suggest that slow-paced mechanics are not just esthetic choices but constitute a design logic aligned with psychological well-being.

Sentiment analysis further confirmed that emotional relief was the dominant affective response among players. Relief-oriented sentiments were especially pronounced in reviews discussing pastoral esthetics and unpressured gameplay experiences. Although some expressions of anxiety were observed, they were comparatively infrequent and often linked to specific in-game challenges rather than general dissatisfaction. This distribution affirms the role of slow-paced design as a supportive emotional environment, rather than one that merely avoids stress.

5.2. Mechanism Interactions and Behavioral Implications

Mechanism tagging and co-occurrence network analysis provided additional insights into the emotional ecosystem of gameplay. Emotional mechanisms such as emotional relief, pastoral healing, and escapism frequently co-occurred, indicating their synergistic roles in shaping positive affective experiences. In contrast, Anxiety Expression appeared more isolated in the network, suggesting that emotionally dissonant experiences were peripheral rather than normative.

Regression analysis linked these emotional mechanisms to observable behavioral outcomes. Emotional relief and escapism were significant predictors of a longer playtime, even after controlling for variables such as review length and recommendation status. This suggests that emotional resonance, rather than traditional game performance factors, is a key driver of sustained engagement. Interestingly, the slow-paced gameplay label itself did not remain a significant predictor once emotional factors were included, implying that pacing operates as a contextual enabler whose effect manifests through emotional mechanisms rather than as a direct motivational force.

The weak explanatory power of the models (R² = 0.015 − 0.036) suggests that individual-level variables (e.g., personality traits, mental health history, or gameplay motivation) likely mediate behavioral outcomes. Nonetheless, the statistical significance of core emotional mechanisms strengthens the claim that affective resonance, rather than challenge-based engagement, plays a pivotal role in sustained gameplay within this genre.

5.3. Summary of the Key Findings and Future Directions

This study addressed the research questions by demonstrating that (1) players express both anxiety and emotional relief, with relief being predominant; (2) slow-paced game mechanics and pastoral esthetics are closely associated with positive emotional sentiment; and (3) emotionally resonant mechanisms predict a longer playtime, reflecting deeper behavioral engagement. These findings highlight the potential of emotionally supportive design strategies in enhancing the user experience within digital environments.

Several limitations of this study should be acknowledged. First, the dataset was exclusively sourced from Steam user reviews, which may introduce selection bias by overrepresenting specific player demographics or gaming experiences. Second, although sentiment classification and mechanism tagging were supported by coherence and performance metrics, the use of automated methods—particularly GPT-based tagging—may still introduce classification bias or interpretive limitations. Moreover, the relatively low variance explained by the regression models suggests that key behavioral or psychological predictors may have been omitted. Additionally, the analysis focused only on reviews written in English and Chinese, which comprised the majority of the dataset. Reviews in other languages were excluded due to the limited sample size and concerns about translation consistency. Future research could address these limitations by incorporating a broader range of multilingual data, supported by robust machine translation pipelines or native multilingual sentiment models, to enhance cultural inclusivity and capture a wider spectrum of emotional expressions.

Future research should pursue longitudinal or experimental designs to capture how emotional engagement evolves over time. Cross-title comparison and physiological validation (e.g., biometric feedback or stress biomarkers) would provide more direct evidence of therapeutic value. Additionally, integrating user-centered design studies could clarify how emotional affordances are perceived and operationalized across different player groups. By expanding both the methodological rigor and theoretical grounding, future studies can further elucidate the role of emotionally supportive gameplay in digital well-being ecosystems.

6. Conclusions

This study investigated how slow-paced game mechanics and emotionally supportive design in Stardew Valley influence players’ emotional states and behavioral engagement. By integrating topic modeling, sentiment analysis, mechanism tagging, and regression modeling, we found consistent evidence that the game’s tranquil mechanics and immersive esthetics promote emotional relief and increased player engagement. This work represents one of the first large-scale empirical validations of slow-paced digital games as potential tools for emotional regulation, addressing a significant gap in the literature on digital health technologies and stress management.

• Our findings suggest that low-pressure, slow-paced gameplay predominantly evokes emotional relief, with dominant affective themes related to calmness, routine, and immersive pastoral settings. Regression analysis confirmed that mechanisms such as emotional relief and escapism significantly predict an increased playtime, underscoring the behavioral relevance of emotional design. Moreover, the game’s structure aligns with SDT [13] by fulfilling psychological needs for autonomy, competence, and relatedness, thus supporting its efficacy as an emotionally supportive environment.
• Methodologically, this study contributes a novel multi-layered approach to investigating affective mechanisms in digital play, integrating large-scale NLP-driven annotation with behavioral outcome modeling. This computational framework is transferable to other game titles and interactive environments, offering a systematic tool for exploring affective engagement in digital media.
• From an applied perspective, the results offer actionable insights for the design of emotionally aware interactive systems. Developers may enhance the emotional support by embedding features such as customizable environments, self-directed pacing, and non-punitive goal structures. Mental health practitioners and educational institutions may also consider integrating slow-paced games into stress management and emotional support programs.

We acknowledge certain limitations. The analysis was limited to reviews in English and Chinese, which represent the majority of the dataset. Reviews in other languages were excluded due to the small sample size and translation consistency issues. Future research could incorporate multilingual review data, supported by robust machine translation pipelines or multilingual sentiment models, to broaden the cultural representation and emotional variability.

Future research should also pursue physiological validations (e.g., biomarker studies), longitudinal experiments, and user-centered evaluations to strengthen the causal inferences and expand the therapeutic relevance of slow-paced games. These directions will help advance the application of interactive technologies in the domains of digital mental health, affective computing, and emotionally sustainable design.