Technology and Emotions: AI-Driven Software Prototyping for the Analysis of Emotional States and Early Detection of Risky Behaviors in University Students

Abstract

Technology-assisted emotion analysis opens new possibilities for the early identification of risk behaviors that may impact the well-being of university students, contributing to the creation of healthier, safer, and more proactive educational environments. This pilot study aimed to design and develop a technological prototype capable of analyzing students’ emotional states and anticipating potential risk situations. A mixed-methods approach was adopted, employing qualitative methods in the ideation, design, and prototyping phases and quantitative methods for laboratory validation to assess the system’s accuracy. Additionally, mapping and meta-analysis techniques were applied and integrated into the chatbot’s responses. As a result, an educational technological innovation was developed, featuring a chatbot structured with a rule-based dialogue tree, complemented by an ontology for knowledge organization and a pre-trained artificial intelligence (AI) model, enhancing the accuracy and contextualization of user interactions. This solution has the potential to benefit the educational community and is also relevant to legislative stakeholders interested in education and student well-being, institutional leaders, academic and well-being coordinators, school counselors, teachers, and students.

1. Introduction

The integration of technology into the educational field is a trend that opens new possibilities and broadens the horizons of knowledge. In this context, the development of technological tools based on artificial intelligence (AI) has significantly transformed various educational scenarios, standing out for their ability to analyze, interpret, and respond in a timely manner to the needs and expectations of students. A clear example of this reality is the application of technologies for emotion identification, facilitating a better understanding of the emotional behaviors that influence retention, social development, and academic performance.

In the current context of higher education, students face increasingly complex emotional challenges, such as stress, anxiety (Trunce Morales et al., 2020), and other states that can lead to adverse consequences, including academic dropout, school coexistence issues, risk behaviors, and significant impacts on their mental health (Jiménez Hurtado et al., 2023). Although universities have implemented psychological support and well-being services to address these needs, these efforts are often reactive and limited in scope, making it difficult to identify students’ emotional needs early and implement preventive strategies. Following this reality, it is emphasized that, globally, 168 million adolescents and young people face mental health issues, with 44.7 percent of the population in Colombia experiencing similar challenges (UNICEF Colombia, 2024). Additionally, 230 cases of suicide have been reported within this demographic, highlighting the urgent need to implement mechanisms for the early identification of warning signs, continuous monitoring of student well-being, and the application of personalized interventions. On the other hand, the literature shows significant advancements in the use of emerging technologies, such as artificial intelligence, for emotion analysis. However, these technologies have not yet been integrated into university contexts to facilitate the prevention of risk behaviors. This highlights a significant gap in the research and development of technological solutions that combine personalization and adaptability to address the emotional needs of this population.

Based on the above, this study makes a significant contribution from the perspective of educational technology to address the central issue of the lack of technological tools that enable the identification and self-regulation of university students’ emotional states. The goal is to facilitate the early detection of risk behaviors and promote emotional well-being. Complementing this, the research question is formulated to determine how emerging technologies can contribute to the self-regulation of university students’ emotional states to facilitate the early identification of risk behaviors.

Technology is a key resource for addressing human needs, including the identification and self-regulation of emotions. Its ability to recognize and analyze data with precision and speed enables the prevention of mental health risks, providing effective solutions based on timely and detailed information.This statement is supported by the study conducted by Alslaity and Orji (2022), who treat the development of systems for the detection of emotions through machine learning techniques and infer that these techniques allow for the analysis of large volumes of data to extract emotional information, facilitating more natural and fluid interactions between humans and computers. From the same perspective, Tawsif et al. (2022), in their research on human–computer interaction, propose the use of physiological signals for emotion recognition. This approach not only significantly improves the ability of systems to accurately interpret human emotions but also opens new possibilities for designing more personalized and adaptive interactions that respond to the specific needs and behaviors of users. Additionally, Bilquise et al. (2022) highlighted the incursion of emotionally intelligent chatbots as one of the most innovative applications of advanced techniques of AI and natural language processing. They stand out, as the training of these chatbots not only gives them the ability to detect and analyze the emotions of the users in real time but can also generate emotionally appropriate responses, adapting to the affective state of the interlocutor and providing more empathetic and personalized interactions by marking before and after the evolution of conversational technologies.

Although the recent literature acknowledges the importance of technologies for emotion analysis in educational contexts, a gap is identified, primarily related to the fact that most of these tools do not integrate validated psychological scales or contextualized normative frameworks on school coexistence. Furthermore, few studies have developed solutions that combine emotional analysis with robust and personalized strategies for emotional self-regulation in university settings. This study seeks to address this gap by developing and validating an educational technology that responds to these needs.

This article presents the process of designing, prototyping, and developing an educational technology conceived as an emotional support tool for university students, promoting a more positive academic experience. The proposed innovation stands out for integrating advanced chatbot programming technology with a pretrained AI model. It uses the Positive and Negative Affect Schedule (PANAS) scale, which has been internationally validated to measure two dimensions of emotional states and the premises of the Colombian National School Coexistence System to recognize possible risk behaviors. In response, the chatbot was trained and validated in the laboratory to offer students recommendations for managing their emotions on the basis of bibliographies from different scientific fields, becoming a technological tool that responds to key challenges related to the emotional well-being of university students in an increasingly demanding context.

This technology contributes to the field of knowledge by integrating an interdisciplinary approach that combines technological advancements, psychological theories, and specific normative frameworks on school coexistence. It provides new perspectives for educational research and the design of solutions aimed at emotional management and student well-being while establishing a solid foundation for future applications in higher education. The article integrates the contextualization of innovation and a theoretical framework that emphasizes the incursion of technology in education, the automatic detection of emotions through technology and recommendation systems, and risk behaviors in students. It exposes the method of design, prototyping, development, and validation of the innovation. The results obtained from the processes developed are presented, and these results are discussed from a theoretical basis. Finally, the article concludes with the main aspects of the study, highlighting the benefits and limitations of innovation and presenting recommendations for the development of future innovations.

2. Foundations from the Literature

2.1. Research on Educational Technology

The technological revolution in education driven by the advances of the 5.0 era has caused an unprecedented transformation in the landscape of teaching and learning, opening doors to horizons and possibilities that were never thought of. Bozkurt (2020) highlighted that research on educational technology has evolved significantly since 1993, moving from approaches based on multimedia and instructional design to the use of big data and intelligent technologies. This evolution reflects a growing interdisciplinarity and greater participation of developed countries in research. Yildiz et al. (2020) inferred that educational technology is the operationalization of scientific knowledge produced in the sciences of education and its application in practice. They identify in their study that current trends in research on educational technologies focus on the practical application of scientific knowledge, emphasizing the importance of adapting educational technologies to the specific needs of students and teachers. Another perspective is presented in the study by Gusho et al. (2023), who indicate that research on educational technology has focused on improving the learning environment, developing effective pedagogical approaches, optimizing assessment, and providing support to students with special needs. They emphasize that the possibilities of educational technologies have been expanded to facilitate access to educational resources, design personalized teaching methods, and improve feedback and evaluation for all students. In this context, research on educational technologies and their development in the 5.0 era should be continued and deepened to identify both existing gaps and new possibilities. This will allow educators and students to adapt effectively to technological innovations, allowing them to use most of these tools to enrich their educational process and better prepare for future challenges.

In line with this evolution, students, as primary actors, have found it necessary to adapt and advance along with educational technologies. They have had to learn to use them effectively to take full advantage of their benefits and to be able to access these tools from anywhere, thus facilitating their training process. Castro (2019) emphasized that digital tools in learning not only facilitate access to education for a greater number of students but also personalize learning activities, improving academic performance and student satisfaction. These tools allow them to learn at their own pace more efficiently, contributing to a more inclusive educational experience tailored to individual needs. Authors such as Manzano Pérez et al. (2023) noted that technological innovation in education not only strengthens the quality of teaching but also promotes self-learning, develops critical skills, and fosters a more interactive and engaged learning environment. In line with this perspective, Bedenlier et al. (2020) stated that educational technology significantly supports student participation in higher education. They highlight the importance and effectiveness of tools such as blogs and mobile learning to promote student engagement, facilitating more dynamic and enriching interactions in the educational process. Together, these technologies not only enhance the active participation of students but also contribute to creating a more inclusive and accessible learning environment. They facilitate the implementation of innovative and effective pedagogical strategies for teachers and provide researchers with valuable opportunities to explore new methodologies and optimize the use of digital tools in education.

2.2. Automatic Detection of Emotions Supported by Technology and Recommendation Systems

Detection and emotion analysis has become a valuable strategy for preventing and detecting possible behaviors that may lead to risky situations, especially in educational settings. This approach allows educational institutions to identify early signs of stress and emotional distress, intervening before situations evolve and turn into crises. Rodríguez-Riesco et al. (2022) reported that the use of technologies for emotional analysis can reveal patterns of anxiety and depression that are not easily detectable through traditional observation and evaluation methods. Similarly, Ringeval et al. (2013) emphasized that the integration of AI systems in emotional evaluation has been shown to be effective in identifying subtle changes in the emotional state of individuals, providing an early warning for intervention. It is important to highlight that the ability of these systems to process large volumes of emotional data in real time enables continuous and accurate monitoring, facilitating timely and appropriate intervention in risk situations (McDuff et al., 2015). These details account for the potential of technology and how it has been integrated into emotional analysis, allowing a deeper, faster, and more accurate understanding of human emotions.

The design and development of technological tools for the automated detection and analysis of emotions are currently a central concern in educational research. In this context, multimodal approaches, which integrate the analysis of emotions through text, sound, images, video, and physiological signals, offer a more advanced spectrum of possibilities. Taken to the educational field, these multimodal methods allow not only for a deeper understanding of the emotional state of the student but also for the faster and more precise recognition of said emotions, anticipating the identification of possible risky behaviors (Marechal et al., 2019). The identification and management of emotions through text analysis with machine learning represents a significant challenge in the implementation of technological and scientific advances aimed at improving the interaction between students and machines. Chatbots, also known as chatterbots, are products of AI that are capable of integrating with any messaging application. These programs can hold conversations with humans via voice commands, text, or both and are considered among the most promising applications of natural language processing (Machová et al., 2023). The literature classifies and describes chatbots in different categories. One of the most common forms of classification is according to the communication channel, which can be text, voice, or images. From the perspective of the knowledge domain, that is, the scope and specialization of the information that they handle, generic or open/cross-domain chatbots, which operate in more than one area of knowledge, and domain-specific chatbots or closed chatbots, which are limited to answering only questions related to a particular knowledge area, are distinguished. The objectives or purposes they seek to fulfill can be divided into informative and conversational chat and tasks (Adamopoulou & Moussiades, 2020). The ability of conversational chatbot systems to maintain a natural conversation with the user, simulating the interaction of a real person, is determined by the design of the conversational workflow, the use of advanced techniques of natural language processing (NLP), and the implementation of a well-structured ontology. The ontology provides a formal representation of knowledge in a specific domain, which allows the chatbot to better understand the relationships between concepts and generate more precise and coherent responses. The response generation module is based on this ontological structure to produce responses via one or more of the three main response approaches, which are the following: rule-based models, which follow predefined patterns and take advantage of the ontology to ensure that the rules are aligned with the knowledge of the domain; recovery-based models, which select the most appropriate response from a knowledge database that is organized and structured by ontology; generation-based models, which create original responses in real time through neural networks and deep learning and use ontology to ensure that the answers are contextually correct and semantically relevant (Colace et al., 2018)

In the educational field, the incorporation of technologies, especially chatbots, powered by machine learning is presented as an innovative alternative both for academic activity and to support the emotional well-being of students. In higher education, chatbots are used in various support roles, both as service assistants and in teaching roles. Its implementation not only aims to improve the academic aspect but also aims to promote accessibility to support services, promote cultural inclusion, and detect the mood of students, thus representing a significant advance in the integration of educational technology (Pérez et al., 2020). An example of the above is the study by Senapati and Phapale (2023), which evaluated the implementation and testing of an emotional support chatbot, yielding promising results for education. Among the total number of students who used it, 85 percent favorably valued it, and those who used it regularly reported lower levels of stress and anxiety, along with an increase in positive emotions such as confidence and happiness. Consequently, 82 percent said that they understood and 70 percent experienced an improvement in their emotional well-being. Another relevant example in the educational field that highlights the integration of emotional support chatbots is EduChat (Dan et al., 2023), a large-scale chatbot system based on language models designed to offer personalized and intelligent education. Among its multiple capabilities, EduChat stands out for offering emotional support to students, owing to its training in a variety of rules and datasets. This chatbot works as an empathetic and compassionate advisor on the basis of a psychological research framework that incorporates approaches such as Rational Emotive Behavioral Therapy and other methodologies focused on the psychology of emotions. The implementation of these conversational tools, with their ability to generate precise, coherent, and contextualized responses, has the potential to positively influence educational processes, facilitating more personalized, empathetic, and accessible learning and the early identification and management of the emotions of students.

2.3. Emotions and Risky Behaviors in Students

Emotional regulation in university students is a continuous and essential process for overall well-being and the prevention of risky behaviors, as the inadequate management of emotions can lead to impulsive and potentially dangerous behaviors. The relationship between emotions and risky behaviors in university students is a topic of growing interest in research. According to Hatkevich et al. (2019), difficulties in emotional regulation during adolescence are strongly associated with the development of severe psychiatric disorders, substance abuse, and suicidal behaviors. They argue that the inability to adequately manage negative emotions can lead to risky behaviors as mechanisms to cope with emotional distress. Similarly, Cerniglia et al. (2023) emphasize the importance of emotional regulation as a central factor in preventing dangerous behaviors during adolescence. These authors examined the interrelations between risky behaviors and difficulties in emotional regulation, finding that the lack of access to emotional regulation strategies is linked to self-harm and risk-taking behaviors. A study conducted by Trógolo et al. (2022) supports this idea by demonstrating that adolescents who experience intense negative emotions and have difficulties regulating them tend to adopt impulsive behaviors as a way to mitigate emotional distress. The authors note that although these behaviors may provide temporary relief, but, in the long term, they worsen mental health problems, perpetuating a negative cycle that affects both the psychological and physical well-being of young people. Together, the studies highlight how the inability to adequately regulate emotions can lead to potentially dangerous behaviors as escape mechanisms. When young people lack effective tools to manage their emotions, they are more likely, in an attempt to relieve emotional distress, to resort to impulsive decisions that can be classified as risky behaviors, affecting both their physical and mental well-being.

Risky behaviors are behaviors that increase the likelihood of negative consequences for individuals and society; these behaviors can include substance use, unprotected sexual behaviors, violence, suicide attempts, and eating disorders. Adolescents are especially vulnerable to these behaviors because of the rapid changes they experience in their physical, cognitive, and social development. These behaviors have the potential to cause physical, emotional, or social harm, affecting both the individual and the community in general (Leather, 2009).

According to Sanci et al. (2018), risky behaviors are characterized by a high potential for physical, social, or emotional problems. During adolescence, individuals are more likely to engage in these behaviors due to various factors, including hormonal changes, social pressures, and the search for identity. Studies have shown that these behaviors are not the result of a single factor but rather of a complex interaction of elements, such as peer influence, a lack of family support, and adverse personal experiences, which make adolescents more prone to take risks.

From an evolutionary perspective, risky behavior in adolescents can also be interpreted as a strategy to obtain potential benefits related to survival and reproduction. This approach suggests that some risky behaviors are not simply impulsive but can be strategically planned. This view challenges traditional frameworks that attribute adolescent risky behavior primarily to a lack of impulse control (Maslowsky et al., 2019). For example, engaging in activities that challenge boundaries can be a means of gaining respect and acceptance within a social group, which is vital during adolescence.

The environment in which adolescents operate also plays a relevant role in the manifestation of risky behaviors. Environmental influences, such as access to substances, exposure to violence, and cultural norms, can encourage or deter the development of these behaviors. Prevention and education programs can be effective in addressing not only individual behavior but also the environmental conditions that contribute to risky behaviors.

University students face a number of challenges that can contribute to the adoption of risky behaviors. These factors include the transition to a new social environment, lack of parental supervision, academic pressure, and the desire to belong to social groups. Experimentation and sensation seeking are common characteristics at this stage of life and can lead to behaviors that put physical and mental well-being at risk (Leather, 2009). Furthermore, El-Ansari et al. (2009) noted that the presence of multiple risky behaviors in college students is associated with an increased risk of poor academic performance, morbidity, and premature mortality. Studies have shown that behaviors such as smoking, excessive alcohol consumption, and physical inactivity are prevalent in this population. These habits not only affect the immediate health of students but can also have long-term consequences, such as chronic diseases and difficulties in professional life. The pressure to excel academically and social expectations often exacerbate these problems, leading students to seek relief in unhealthy behaviors.

To address these challenges, it is necessary to implement preventive strategies that do not focus solely on a single risk factor but that address multiple possibilities. Universities can play a critical role in providing support programs that promote healthy lifestyles, as well as counseling and mentoring services to help students manage stress and anxiety. It is also important to foster an inclusive and supportive environment where students feel safe to seek help without fear of being judged; promoting recreational and sports activities offers them healthy alternatives that help them channel their energy in a positive way.

3. Materials and Methods

This study follows a mixed-methods research approach and has been conceived as a pilot study, predominantly integrating qualitative methods in the ideation, design, and prototyping phases to understand key needs and perceptions for the development of a prototype. The method applied in this phase is primarily related to the Design-Based Research methodology, which combines empirical educational research with theoretical design in learning environments. This approach aims to create and evaluate educational interventions in real-world contexts, adjusting the design based on observations and feedback received (Baumgartner et al., 2003). In the laboratory validation phase, quantitative methods were employed to assess the system’s accuracy in a controlled laboratory environment. Additionally, mapping techniques and meta-analysis were applied to identify and extract the main self-regulation recommendations generated by the system.

The objective of this study was to design and develop a technological prototype aimed at analyzing the emotional states of university students to offer support and facilitate the early detection of possible risk behaviors. For this purpose, software programming technology and a pretrained AI application capable of analyzing and processing emotional data in real time were used. The research question that guided this study was as follows: How can emerging technologies contribute to the management of the emotional states of university students, facilitating the early identification of risk behaviors? Next, the stages followed in the ideation, design, prototyping, development, and laboratory validation of the innovation are detailed at a methodological and practical level.

3.1. Ideation, Design, and Prototyping Methods

The method used for ideation, design, and prototyping is based on participatory and collaborative approaches characteristic of qualitative methodology, with a particular emphasis on User-Centered Design (UCD). This design approach was implemented in two stages as follows:

• Cocreation: In this initial stage, 50 university students from different semesters of the academic program of risk management, safety, and health at work, belonging to a Colombian public higher education institution, were integrated as active agents of the design process. Those who gave their consent to participate in a collaborative group activity conducted in a single session were distributed across 7 randomized work groups; the students identified priority areas of intervention linked to their needs for support and student accompaniment, providing key perspectives to guide the development of the prototype and promoting a user-centered approach aligned with their particular contexts. The cocreation session made it possible to collect various perceptions, identify relevant problems, and validate in a contextualized way the proposals for the improvement of the support and accompaniment processes.

  Table 1. Identification of perceptions and areas of intervention.

  Working Group No.	Students’ Perceptions	Priority Intervention Areas
  Group 1	Insufficient opportunities or spaces to interact and build relationships with the support staff, limiting communication and personalized support.	Create spaces where students can express themselves creatively, with workshops, activities, and initiatives that encourage creativity and free expression.
  Group 2	Lack of spaces for relaxation and disconnection from academic responsibilities, limiting emotional well-being and balance on the university campus.	Develop flexible learning environments that allow students to experiment with different ways of learning and collaborating, adapting to their individual learning styles.
  Group 3	Insufficient support and physical resources to handle the daily challenges and problems of the university environment.	Implement an accompaniment program that guarantees protection and adequate guidance in various situations.
  Group 4	Low offer of extracurricular activities that respond to student interests and needs, limiting participation and sense of belonging.	Design an inclusive accompaniment program that responds to the different interests and needs of students.
  Group 5	Lack of clear and accessible guidance to facilitate decision-making and adaptation to the campus, generating confusion and disorientation when making informed academic and personal decisions.	Establish an accompaniment program that provides support in the academic and personal trajectories of students.
  Group 6	Lack of specialized counseling and psychological support services to effectively manage mental and emotional well-being in situations of stress or difficulty.	Incorporate an emotional support component in the accompaniment program, which allows for the identification and monitoring of emotions and feelings.
  Group 7	Weak recognition of individuality and diversity; students are not valued or recognized for their cultural, social, or academic differences.	Develop an accompaniment program that values inclusion and diversity in all its dimensions, namely cultural, social, and academic.

  presents a summary of these perceptions as well as the improvement proposals made by the participating students.
• Prototyping: Using simple stationary materials, each group of students developed prototypes, physical models, or mockups aimed at proposing viable solutions to address the identified priority areas of intervention. These prototypes were subjected to a validation process among the students themselves, allowing direct feedback to be obtained and thus enriching the initial proposals.

In the collaborative workshop, which applied the User-Centered Design method (see Figure 1), the idea with the greatest impact was selected through an interactive vote. The students, using their mobile devices, evaluated and chose the proposal they considered the most appropriate, on the basis of criteria such as feasibility, potential impact, and coherence with the priority areas of intervention. As a result, the students selected the idea of a “humanoid robot” capable of monitoring students’ emotions and sending automatic alerts to offer emotional support in real time.

In general terms, as Miranda et al. (2020) noted, the collaborative process promoted and facilitated the generation and socialization of ideas among students, fostering a creative, participatory, and conscious approach to meeting the needs and requirements of university students.

3.2. Educational Technology Development Method

Upon concluding and documenting the results of the collaborative workshop, and on the basis of the prototype selected by the students, the group of researchers channeled and refined the concept, transforming it into an educational technology based on programming software supported by AI. The innovation technical development process was structured according to the phases shown in Figure 2.

3.3. Initial Validation Method in the Laboratory

Once the educational technology (chatbot robot) was developed, a validation process was carried out to determine the level of accuracy in identifying emotions, according to the affective states of the PANAS scale, validated for a sample of university students in Colombia (Moreta-Herrera et al., 2021). The validation of the chatbot was conducted using pre-designed test cases based on the risk typologies analyzed by the robot (Kaner et al., 1999) before and after the integration of the AI “RoBERTuito” using cases. The validation followed the following phases:

• Use case design: Use cases were documented for each risk category (Types I, II, and III), simulating critical and relevant scenarios for the university context.
• Assignment of emotional dags: A specific emotion tag was assigned to each predefined case on the basis of the emotion identified in the content.
• Preintegration accuracy validation: Chatbot accuracy was evaluated prior to AI integration.
• Postintegration accuracy validation: Chatbot accuracy was evaluated after AI integration to analyze possible performance improvements.
• Registration and optimization: The results of the validation were documented, identifying areas of improvement in the detection of emotions and responses generated by the chatbot.

To enhance the validity of the research design, the precision validation record file Validation Document (https://docs.google.com/spreadsheets/d/1oSJt4gmjbxREIT0hgz4AceCV4Klcw8wN/edit?usp=sharing&ouid=105191236480538870852&rtpof=true&sd=true, accessed on 27 February 2025) presents the cases or scenarios evaluated for each risk level.

4. Results

In the presentation of the results of this study, several aspects are addressed to understand how the initial concept was transformed into a technological reality. The innovation is detailed from its conception to its technical development and laboratory validation, highlighting the operational functionality of the chatbot, as well as the technologies and tools used in its development. Additionally, the training program planned for the deployment stage is described, aimed at both end users and administrators of the technological tool, including report generation and monitoring.

4.1. Description of the Innovation: From Concept to Technological Reality

Taking as a reference the perceptions identified during the co-creation and prototyping process collected in the collaborative workshop, the researchers developed an emotional analysis and management robot called DOCO, an acronym in Spanish for “Docente Consejero” (Advising Teacher). This robot has the ability to capture and analyze information about university students’ emotions in real time, identify and measure them, and offer personalized recommendations and support resources for their management. Similarly, it can identify potential risk situations for both the student and the institution.

Students can access the DOCO robot mobile application from their personal devices or use it on interactive screens strategically located in high-traffic areas of the university campus. Based on the acceptance of the personal data processing policy, students provide information such as age and academic level, among others. Next, the student self-reports his state of mind via the emotions defined in the Positive and Negative Affect Schedule (PANAS) evaluation scale, which was proposed by psychologists David Watson, Lee Anna Clark, and Auke Tellegen (Watson et al., 1988). This scale, widely used in the fields of psychology and education, makes it possible to reliably measure 20 emotions in two main dimensions: positive affect and negative affect. The PANAS scale integrated into the DOCO robot corresponds to the PANAS scale validated in a sample of university students from Colombia and Ecuador (Moreta-Herrera et al., 2021) (

Table 2. Positive and Negative Affect Schedule (PANAS) assessment scale.

Dimension	Emotions
Positive affect	Enthusiastic, Interested, Determined, Excited, Inspired, Alert, Active, Strong, Proud, Attentive
Negative affect	Scared, Afraid, Upset, Distressed, Jittery, Nervous, Ashamed, Guilty, Irritable, Hostile

). This scale, which is based on the reliability and factorial invariance of the PANAS scale in both countries, ensures that the evaluation of students’ emotions is accurate and culturally relevant. The information provided by students is processed by the robot to offer support and emotional support tailored to their specific emotional needs in an immediate and personalized way.

The management recommendations provided by the robot are supported by principles from educational, psychological, neuroeducational, medical, and other fields, ensuring a solid, reliable, and objective operational foundation. These recommendations have been compiled from the results of a systematic literature review on emotional self-regulation within the mentioned fields of knowledge, using indexed databases (Scopus, PubMed, Web of Science, and PsycINFO) and the peer-reviewed academic literature. The review included studies exploring theoretical approaches, practical interventions, and innovative technologies applied in educational and clinical contexts, enabling the identification of evidence-based emotional regulation strategies. Furthermore, these recommendations have been carefully reviewed to ensure their applicability to diverse educational contexts, including those with high emotional vulnerability.

The search was conducted using the following search string: (“emotional regulation” OR “emotion management” OR “self-regulation”) AND (“educational settings” OR “schools” OR “higher education”) AND (“psychological interventions” OR “neuroscience” OR “mental health” OR “student support”) in indexed databases such as Scopus, PubMed, Web of Science, and PsycINFO. The number of included documents, after applying inclusion and exclusion criteria, is summarized in

Table 3. Systematic review records report.

Stage	Description	Number of Records Processed
Identification	Records identified in Scopus	18,000
	Records identified in PubMed	10,000
	Records identified in Web of Science	12,000
	Records identified in PsycINFO	8000
	Total records identified	48,000
	Records removed as duplicates	13,800
	Records after duplicates removed	34,200
Selection	Records excluded after title and abstract review	28,000
	Records selected for full-text review	6200
Eligibility	Records excluded after full-text review:
	Lack of focus on technologies or emotional self-regulation	3000
	Previously undetected duplicate studies	1000
	Publications in unselected languages	200
Inclusion	Included in qualitative synthesis	1200
	Included in meta-analysis	800

. This table outlines the stages of the selection process, from the initial identification of records to their final inclusion in the qualitative synthesis and meta-analysis.

DOCO collects data on the emotions of students through the interactions and analyses of patterns. This information is processed in real time to identify signs and behaviors that can be framed as possible risk behaviors, in accordance with the provisions of the National Coexistence System Colombian School (República de Colombia, 2013). The emotions and other data recorded by the students help the robot DOCO understand their emotional state, and the analysis of this information is correlated with the robot, owing to the integration of a model of pretrained artificial intelligence called “RoBERTuito”, which makes it possible to relate the emotions detected with the possible situations that, according to the regulations, may affect school coexistence and the exercise of human, sexual, and reproductive rights. These situations are classified into three types:

• Type I situations include improperly managed conflicts and sporadic situations that negatively affect the school climate but do not damage the body or health.
• Type II situations refer to situations of aggression, bullying, and cyberbullying in school that do not meet the characteristics of a crime and that present any of the following characteristics: *They manifest repeatedly or systematically. *They cause damage to the body or health without generating disability for any of those involved.
• Type III situations include situations of aggression in school that constitute alleged crimes against freedom, integrity, and sexual training, in accordance with the provisions of the current criminal law in Colombia.

Given that one of the proposed objectives of the technology development is to provide a proactive support system that also enables student support staff to intervene before emotional issues escalate into crises that could affect the student or the educational community, the robot has been programmed to generate alerts directed to administrators and student counselors. This allows for a swift and appropriate response to students’ emotional needs.

Figure 3 graphically displays the start window, including the robot’s conversation interface. Its intuitive, user-friendly, and accessible design stands out, facilitating navigation and quick access to the main functions, ensuring an optimized and efficient user experience for students.

Similarly, to provide greater clarity regarding the robot’s interactions with the student, Figure 4 shows an example of the dialogue generated by the robot in the scenario of registering positive emotions. In this case, two variations are included: when the student wishes to receive recommendations to maintain these emotions, and when they choose not to request them. This resource aims to illustrate how the system identifies, interprets, and adapts its responses based on the student’s needs and preferences, promoting both self-regulation and proactive emotional support. Additionally, the example highlights the robot’s ability to respect the student’s autonomy while fostering a personalized and effective interactive experience.

4.2. Conceptualization and Technical Development

This chatbot is structured as a dialog tree, a common technique in rule-based chatbots that guides interactions through predefined flows, where each node represents a predefined interaction or question. User responses trigger the transition to a different node within the tree. This ensures that conversations follow a defined and controlled logic.

• Tree structure: The dialog tree is organized hierarchically, with nodes representing possible user inputs and system responses. Decisions within the tree are based on defined rules (transition rules), through which logical conditions that determine the next node on the basis of the identified emotion have been implemented.
  The dialog flows are designed to go from general questions to more specific questions on the basis of the user’s responses, thus allowing a detailed exploration of the user’s emotional state.
• Design and development of the rule-based chatbot: The application was developed in its user interface (front-end) via Blazor, an open-source framework from Microsoft, which guarantees a fluid and adaptable user experience. The mobile version, compatible with both Android and iOS, was created using .NET MAUI, a tool that allows the development of applications for multiple platforms, integrating the C and XAML programming languages for interface design. On the back end, the .NET Core 8 API was used, with the data being stored in an SQL server database, which ensured efficient and secure information processing.
• Integration of the AI Model into the Chatbot: A pre-trained AI model has been integrated into the rule-based chatbot (dialogue tree), which uses mapping rules to identify language patterns associated with the emotions from the PANAS model and the consequent risky behaviors as outlined in Colombian Regulatory Decree 1965 (República de Colombia, 2013). This AI model, named “RoBERTuito,” acts as a complement to the system, analyzing user responses to detect specific patterns corresponding to positive or negative emotional states.
  The AI integration is based on the fact that RoBERTuito is a model implemented in the cloud and accessible via a RESTful API. This means that the chatbot sends the text entered by the user to RoBERTuito’s API and receives a response containing the emotion classification and a confidence level.

RoBERTuito has been used as a pretrained AI model with mapping rules to identify language patterns associated with the specific emotions of the PANAS model. The fine-tuning process is carried out with a labeled dataset that contains affirmations. These affirmations are classified according to their emotional polarity: positive, negative, or neutral. The key aspects of the model training process are as follows:

• Labeled dataset: A dataset of texts in Spanish in which the sentences are already manually classified into categories of emotions was selected. The source for the construction of the data sheet came from social networks (namely X and Facebook), product reviews, or news articles.
• Fine-tuning the model: RoBERTuito adjusted itself via these emotion labels, optimizing its weights so that it learned to predict the emotional polarity of a text on the basis of the context. During this process, the model parameters were updated to maximize the accuracy of the emotion classification.
• CLS classification tokens: The model used a special CLS token at the beginning of each text sequence, which was used to condense the representation of all the text. In the analysis of emotions, this token was essential for the model to emit its final prediction (positive, negative, or neutral) on the basis of the contextualized representation of the text.

4.3. Laboratory Precision Validation Results

The validation was carried out in the laboratory by the design team, following the previously defined validation method. The adaptability of the DOCO robot to respond appropriately according to the emotion identified and the type of associated risk was verified before and after the integration of the AI, according to a reference metric set at 90 percent coincidence.

• Evaluated Scenarios:

• Type I: Minor conflicts and sporadic situations: 20 cases
• Type II: School assault or bullying without crime characteristics: 20 cases
• Type III: Situations constituting serious crimes: 20 cases

To validate the greater precision of the chatbot after the integration of the AI “RoBERTuito”, the results showed that in cases I and II, the precision increased with the integration of the AI, whereas in case III, both systems maintained the same level of precision (95 percent). Below is a graphical visualization that demonstrates the differences in performance for each category of case Figure 5.

The consolidated results by case and risk level are recorded in a spreadsheet, which is available for consultation at the following link: Validation Document (https://docs.google.com/spreadsheets/d/1oSJt4gmjbxREIT0hgz4AceCV4Klcw8wN/edit?usp=sharing&ouid=105191236480538870852&rtpof=true&sd=true, accessed on 27 February 2025).

4.4. User Training

The training process planned for the robot deployment stage is structured in two stages, according to the experience-based training method experiential learning (Chen et al., 2022; Datta, 2023) proposed by David Kolb. This method is based on an experiential learning cycle that includes four stages: concrete experience, observational reflection, abstract conceptualization, and active experimentation.

In the first formative stage, permanent training is developed that includes explanatory material and individual accompaniment sessions, allowing students to reflect on the importance of mental health care and to become aware of this issue. Throughout this phase, users have their first interaction with the robot, allowing them to experiment and explore its functionalities. This direct interaction constitutes the necessary experience for users to become familiar with the pilot implementation program. During this process, users are encouraged to reflect on their experiences to gain a deeper understanding of how the robot can contribute to the generation of emotional well-being.

The second stage will focus on final training, where users receive more detailed and practical instruction on the full use of the robot, both in its mobile version and the fixed version on the university campus. In this phase, experiences and feedback are collected and analyzed to make final adjustments, ensuring it effectively meets the students’ needs. This final training ensures that users are fully prepared to integrate the robot into their routines, maximizing its positive impact on the university community.

4.5. Administration of the Technological Tool

The administration of the technological tool is a shared responsibility that involves various actors within the institution. This collaborative effort focuses on the area of student well-being and the academic vice-rectory, which coordinate the area of student guidance and accompaniment. On the other hand, student counselors and psychologists are important actors in this process, as they can use the platform to proactively monitor and support the emotional well-being of students, ensuring that each receives the necessary support for their well-being emotional development. The management of this software not only increases the capacity of the institution to respond effectively to the emotional needs of its students but also, as highlighted by Aithal and Aithal (2023), strengthens collaboration between the different areas of the university. With the effective coordination between the area of student welfare, the academic vice-chancellor, and the counseling services, the educational institution can offer more cohesive and proactive support, which responds not only to the individual needs of the students but also to the emerging demands of the academic environment.

The main reports generated by the software in relation to the expectation of use are detailed in

Table 4. Main reports generated by the software.

Report Identification	Information Integrated into the Report
Participation and Use Reports	- Comparison between groups of students worry, tension, or anxiety
	- Emotional registration frequency
	- Interaction with support resources
Emotional Well-being Reports	- Tendencies of positive and negative affect
	- Emotional distribution
	- Assessment of individual emotional state
Intervention and Support Reports	- Satisfaction of the support received
	- Identification of risk cases
Academic Impact Reports	- Correlation between emotional well-being and academic performance
	- Retention and dropout rate according to emotional state
Evaluation Reports and Continuous Improvement	- Suggestions and user feedback

. This table provides a comprehensive view of each type of report, including its purpose and the information it offers. These reports range from monitoring emotional states and the effectiveness of interventions to evaluating participation in emotional support programs and the impact on academic performance. The ability to generate these reports allows educational institutions not only to monitor the emotional well-being of students on an ongoing basis but also to make informed decisions to promote a healthier academic environment.

4.6. Ethical and Legal Considerations in the Implementation of the Technology

The design of this technology aims to balance students’ ability to self-regulate their emotions with the institutional support necessary to prevent emotional risk situations that could impact their well-being or that of the educational community. This approach aligns with the Mental Health Guidelines for the Higher Education System issued by the Colombian Ministry of National Education in 2023, which emphasize the importance of fostering educational environments that integrate prevention, timely intervention, and respect for fundamental rights. The technological tool enables students to reflect on and become aware of their emotional state, fostering self-regulation and self-care. The alerts generated are exclusively directed to trained professional staff, such as student counselors or administrators, who act in accordance with established protocols to provide the appropriate support.

In this pilot study, the feasibility and preliminary impacts of the innovation are analyzed before large-scale implementation. This process acknowledges and addresses the ethical and legal concerns associated with the collection and use of emotional data, ensuring compliance with privacy regulations and the protection of participants. In this context, the system adheres to the principles of Law 1581 of 2012 on Personal Data Protection, guaranteeing that students or their legal representatives provide informed consent for the processing of sensitive information. During the ideation phase, conducted through a participatory workshop with students, all participants signed an informed consent form specifying the purpose, the use of the collected information, and the confidentiality measures adopted. In the validation phase, since it was conducted in a controlled laboratory environment without direct student participation or the collection of personal data, approval from the institutional ethics committee was not required. However, in future phases of the study, when the system is evaluated with end users, prior ethical review and approval will be sought to ensure compliance with applicable regulations and the protection of participants.

Likewise, the provisions of Colombian Law 2383 of 2024 are observed, promoting the creation of safe and healthy educational spaces while prioritizing privacy, confidentiality, and human dignity at all stages of data management. In this way, the proposed technology is positioned as a preventive support tool, designed to complement mental health strategies and enhance institutional capacity to proactively address students’ emotional needs.

5. Discussion

The discussion scheme focuses on how research in educational technology, the automatic detection of emotions supported by technology and recommendation systems, and risk behaviors in students are interrelated and how a chatbot (robot) designed to identify and manage emotions can have a significant effect on the prevention of risky behaviors.

The development of a chatbot robot to assist in the identification and management of emotions in students finds a solid theoretical foundation in the evolution of research in educational technology. As highlighted by Bozkurt (2020) and Yildiz et al. (2020), educational technology has progressed toward a more practical and interdisciplinary approach, where the application of intelligent technologies such as AI has become instrumental in enhancing learning. This advancement has facilitated not only the personalization of educational activities but also the ability to address the emotional and psychological needs of students through advanced tools trained for emotional analysis.

From the perspective of educational technology, the incorporation of chatbots in learning environments also aligns with current trends toward more inclusive and accessible teaching. As suggested by Manzano Pérez et al. (2023) and Bedenlier et al. (2020), these systems not only improve the quality of education and self-learning but also foster a more interactive and engaging learning environment. The ability of chatbots to generate natural and adaptive responses, which is based on a structured conversational model, allows them to play a key role in promoting the emotional well-being of students, an important component in reducing the risk of negative behaviors and facilitating personal development and academics.

A fundamental aspect of the chatbot robot design is that it arises from the identification of the real needs of the students, who are the main actors in the educational process. By putting their concerns and emotions at the center, we ensure that the chatbot not only detects emotional patterns but also responds directly to the specific challenges students face in their academic and personal environments. This approach is aligned with the vision of Gusho et al. (2023), who highlighted the importance of adapting educational technologies to address the individual and specific needs of users.

In this context, the automatic detection of emotions with technologies based on AI, as described in the studies by Rodríguez-Riesco et al. (2022) and Ringeval et al. (2013), has become an important strategy for the prevention and management of risky behaviors. The DOCO robot uses the Positive and Negative Affect Schedule (PANAS) methodology (Watson et al., 1988) to evaluate the affective states of the students; this is a tool that was specifically validated in the university population of the project development area (Moreta-Herrera et al., 2021). This methodology not only provides an accurate measurement of positive and negative emotions but also establishes a solid scientific basis for analyzing emotional states and their possible implications for student behavior.

Risky behaviors in students, such as substance use, depression, and violence, are strongly influenced by emotional and contextual factors. Sanci et al. (2018) and Maslowsky et al. (2019) reported that these behaviors are not the result of a single factor but rather of complex interactions among social, emotional, and environmental influences. The use of a chatbot that can identify and correlate these signals with normative aspects of the educational environment to provide personalized interventions not only supports a proactive approach in emotional education but also aligns prevention strategies with the specific needs of the individual, addressing both the causes and the manifestations of these risky behaviors. As a complement, the robot offers personalized recommendations and support resources drawn from the Rational Emotive Behavioral Therapy (Dan et al., 2023) and scientific literature in the educational, psychological, neuroeducational, and medical fields. This interdisciplinary approach ensures that interventions are not only based on emotional detection and evaluation but also supported by strategies and scientific knowledge that facilitate a more effective response tailored to the individual needs of students.

Importantly, the university environment presents unique challenges that can exacerbate risky behaviors in students. The transition to a new social environment, academic pressure, and the desire to belong to social groups are factors that, according to Leather (2009) and El-Ansari et al. (2009), increase the susceptibility to adopting dangerous behaviors. In this sense, chatbot DOCO can offer emotional guidance and specific resources to students, making it a valuable tool not only for managing stress and anxiety but also for promoting healthier lifestyles and reducing the incidence of risky behaviors.

The robot DOCO is distinguished by its ability to optimize its performance and reach multiple advantages. Integrated development with .NET MAUI allows the application to run smoothly on Android, iOS, and desktop platforms. Additionally, using Blazor to create dynamic user interfaces ensures a smooth and engaging user experience. The precision validation carried out allowed us to evaluate the reliability of the chatbot in the identification of emotions and their corresponding classification according to the type of risk. By simulating cases, the design team observed how the chatbot responded to different predesigned scenarios, adjusting their responses to specific emotions and risk levels, which confirmed their efficacy in the management of critical situations in the laboratory.

The results of the validation carried out showed an improvement in the accuracy of the chatbot after the integration of the AI (RoBERTuito), although some aspects that must be refined before being tested with the end user are identified. This validation not only allows us to demonstrate the positive impact of AI on the performance of the chatbot, but also provides important information to continue optimizing the system to offer more precise and sensitive monitoring that allows us to detect risk behaviors early and support the emotional well-being of students.

The chatbot DOCO system presents various challenges that consolidate it as a tool in constant update and evolution. One of the broadest future possibilities is the integration of multimodal systems, which combine the analysis of different signals, such as text, voice, facial expressions, and physiology, to offer a comprehensive and deep vision of the emotional state of the student, as inferred by McDuff et al. (2015). These multimodal systems allow for a more precise and contextualized evaluation by offering the possibility of analyzing and understanding emotions in real time from multiple data sources. Likewise, the integration of other scales of affective states validated for the context of higher education, as well as a broader spectrum of recommendations for emotional management on a scientific basis, represents a valuable opportunity to increase the precision and depth of the system in the detection of emotions, allowing a more complete accompaniment adapted to the needs of the students. Another aspect is related to the continuous incursion of new AI, which could offer opportunities to complement and strengthen, expanding its adaptation and response capacities.

Similarly, the robot faces certain limitations and challenges that must be considered for its effective implementation. First, there is a need for the constant updating and retraining of the model to maintain its effectiveness. This includes not only incorporating new data that reflect changes in students’ communication patterns and emotional expressions but also adapting to diverse cultural and linguistic contexts, especially in heterogeneous populations. The system also faces technical challenges related to its ability to interpret complex emotions, such as mixed or ambiguous states, and to distinguish between explicitly expressed emotions and those that are implicit or non-verbalized. Additionally, the proper management of students’ emotional data privacy and confidentiality becomes a significant challenge. Even with the implementation of protection and anonymization measures, there is always a risk of data security vulnerabilities, requiring substantial efforts to address this aspect effectively. Finally, the implementation of the robot requires ongoing technical and pedagogical support, including the training of educational staff to interpret and act on the alerts generated by the system.

6. Conclusions

To respond effectively to the needs and requirements of university students and offer comprehensive support that fosters their emotional and academic well-being, it is necessary to fully understand their perceptions, experiences, and expectations. In response to these factors, the design and implementation of support strategies that integrate collaborative learning and technological innovation allow better adaptation to the specific demands of student life.

The incorporation of technological tools such as educational chatbots powered by AI allows us to offer continuous and accessible support to the emotions of students. These systems can monitor emotional states in real time, identify possible risk situations and provide resources and guidance immediately, significantly improving the perception of accompaniment and support. These tools are essentially based on a multidisciplinary approach that integrates scientific knowledge from psychology, education, and technology, ensuring that the recommendations have a solid foundation, are empathetic, relevant, and focused on the comprehensive well-being of the student.

This study suggests that conversational chatbots such as DOCO, structured as a rule-based dialog tree that guides interactions through predefined flows and the integration of pre-trained AI models, have the potential to be beneficial for the identification and management of emotions, as well as for the prevention of risky behaviors in students. Through the integration of the Positive and Negative Affect Schedule (PANAS) and the correlation of emotional states with the risky behaviors defined in the Colombian National School Coexistence System, the DOCO chatbot has demonstrated its potential to anticipate and respond to critical emotional situations effectively and proactively.

The design and development of applications and technological tools is a driving factor for innovation in educational institutions. Rule-based chatbots, which operate by following predefined patterns and use ontology to ensure that their responses align with domain knowledge, can significantly enhance institutions’ ability to monitor and support student well-being, thus fostering an educational environment that values and promotes the integration of innovative technological solutions. Likewise, the incorporation of AI and advanced data analysis systems in the educational field expands the ability of institutions to adapt and evolve in an increasingly dynamic and complex environment.

This study demonstrates the potential of the DOCO chatbot as a technological tool for emotional support in the university environment. Through initial validation in the laboratory, it was found that the integration of AI (RoBERTuito) improves the precision and adaptability of robots in the identification of emotions and possible risk situations. The initial results highlight the importance of incorporating advanced technologies in the design of psycho-emotional support solutions, allowing timely and effective interventions in educational contexts.

One of the main benefits of the chatbot is its ability to provide targeted support for students’ emotional needs. Another important benefit is its accessibility, allowing access through interactive screens located in high-traffic areas on the university campus or via mobile devices. This encourages its use and makes it easier for students to seek support when they need it. The technology also offers an anonymous and safe environment where students can share their feelings without fear of being judged, which is especially important during a stage of life when many may feel social pressure or anxiety.

In general, it is recommended that educational institutions integrate this type of technology as part of a comprehensive student well-being program, ensuring the availability of trained human resources to interpret the alerts generated and respond in a timely manner. Similarly, legislators and regulators, as key stakeholders in the process, should establish regulatory frameworks that support the responsible use of AI-based technologies in education, prioritizing the protection of sensitive data and promoting innovation aligned with human and educational rights. For their part, technology developers must provide regular system updates, incorporating relevant data that reflect changes in students’ communication patterns and emotional states. They should also continuously work on improving the security and accuracy of the model, with special attention paid to its applicability in culturally diverse contexts.

Future research should focus on the development and promotion of innovative technologies in the educational field while ensuring their responsible use. This involves not only advancing the design of tools that address the needs of the educational context but also establishing ethical and regulatory frameworks that guarantee privacy protection, equity in access, and sustainability in their implementation. Likewise, addressing concerns about dependency on technology and proposing policies and practices that strengthen the protection of student confidentiality are vital aspects of this field. This approach will not only ensure a safe and ethical educational environment but also deepen the understanding of emotional learning, contributing to the design of more equitable, inclusive, and effective educational systems.