LumiCHAT: An Expert-Based Mental Health Support Chatbot for Students

LumiCHAT: An Expert-Based Mental Health Support Chatbot for Students

Lowell Jay P. Pabua

College of Information Technology Tagoloan Community College Philippines

Cloyd P. Labininay

College of Information Technology Tagoloan Community College Philippines

Lorenz M. Saldivar

College of Information Technology Tagoloan Community College Philippines

Earl Patrick R. Sepida

College of Information Technology Tagoloan Community College Philippines

Kesha Jade S. Sabanpan

College of Information Technology Tagoloan Community College Philippines

Jan James M. Pada

College of Information Technology Tagoloan Community College Philippines

Nikko A. Perez

College of Information Technology Tagoloan Community College Philippines

Abstract – This study developed and evaluated an expert-based mental health support chatbot designed to provide accessible, confidential, and responsive assistance to students of Tagoloan Community College experiencing emotional distress, academic pressure, and psychological concerns. The system employed Artificial Intelligence and Natural Language Processing through RASA, supported by Laravel and React.js, to deliver empathetic conversations, assess risk levels, recommend coping strategies, and activate referral mechanisms to guidance counselors when necessary. Using the Agile Software Development Life Cycle, the platform integrated key features such as anonymous interaction, bilingual communication, automated appointment scheduling, and secure data handling compliant with the Data Privacy Act of 2012. System testing and usability evaluation using the System Usability Scale confirmed that the application is functional, user-friendly, and effective in reducing barriers caused by stigma, limited access to counseling services, and hesitation to seek help. The results demonstrate that the system enhances institutional mental health support by promoting early intervention, strengthening student-counselor communication, and fostering emotional well-being through a safe, inclusive, and technology-driven platform, contributing to a more responsive mental health support system within the academic environment.

Keywords – mental health chatbot, student support system, artificial intelligence, natural language processing, counseling referral, Agile SDLC, higher education mental health

1. INTRODUCTION

   Mental health challenges among college students have increased due to academic pressure, social stressors, and limited access to professional support [1]. Although counseling services exist in many higher education institutions, they are often underutilized because of stigma, scheduling difficulties, and limited availability of mental health professionals [2]. As a result, many students do not receive timely intervention, negatively affecting their well- being and academic performance [3].

   Recent studies indicate that AI-based chatbots can serve as effective supplementary tools for mental health support by providing anonymous, immediate, and nonjudgmental interactions [4], [5]. However, institutional deployment of such systems requires careful consideration of ethical responsibility, data privacy, and system reliability [6].

   To address these concerns, this study developed LumiCHAT, an expert-based mental health support chatbot for Tagoloan Community College. The system functions as an initial support platform that provides coping guidance, self-care recommendations, and referral to institutional counselors when necessary. LumiCHAT was implemented

   using a web-based architecture with secure backend services and an interactive frontend, incorporating natural language processing to interpret user concerns and deliver expert- guided responses [7].

2. RELATED WORK

   Recent studies have demonstrated the growing role of conversational agents in supporting mental health and educational services. Prior research has focused on chatbot architecture, development platforms, natural language processing (NLP) integration, and evaluation strategies relevant to sensitive domains such as mental health [8], [9].

   1. Chatbot Architecture and Development Approaches Agile-based development models are widely adopted in

      chatbot systems due to their flexibility and emphasis on iterative refinement. Skuridin and Wynn proposed an operational framework that links strategic objectives with Agile practices, highlighting the importance of continuous user feedback and rapid prototyping [8]. Similarly, Alegado et al. demonstrated the effectiveness of Agile SDLC in an academic chatbot system, reporting improved adaptability and usability through iterative testing [9].

      Advanced chatbot architectures have also incorporated knowledge-driven structures. Singla et al. introduced CAMI, which employed a knowledge graph to generate context- aware responses, enabling more precise and personalized recommendations [10]. These architectural strategies informed the modular and extensible design adopted in LumiCHAT.

   2. Development Tools and Platforms

      The choice of chatbot development platform significantly affects scalability, transparency, and data control. Pérez-Soler et al. compared popular platforms such as Dialogflow, IBM Watson, and Rasa, concluding that open-source frameworks provide greater flexibility and deployment control [11]. Abdellatif et al. further highlighted that Rasas modular architecture and customizable NLU pipelines make it suitable for research-oriented and privacy- sensitive applications [12]. These findings support the selection of open and locally deployable chatbot frameworks for institutional environments.

   3. NLP and AI in Mental Health Chatbots

      NLP and AI techniques enable chatbots to recognize user intent, analyze sentiment, and generate appropriate responses. Zhou et al. demonstrated that emotionally aware conversational systems, such as XiaoIce, can sustain meaningful user engagement through empathy-driven interaction metrics [13]. Large language models have further improved conversational depth and contextual understanding, reinforcing their suitability for mental health support applications [10].

   4. Applications in Education and Mental Health

      Several studies have confirmed the effectiveness of chatbots in educational and mental health contexts. CAMI successfully delivered evidence-based guidance to caregivers through a participatory design approach [10]. In

      academic settings, chatbots have been shown to influence student decision-making and trust in AI-assisted systems [14]. Locally, Cuevas et al. demonstrated the feasibility of NLP-enabled chatbots in Philippine public service systems, indicating strong user acceptance of conversational platforms [15].

   5. Evaluation Strategies

      Chatbot evaluation commonly combines usability testing, performance metrics, and quality models. Prior studies employed ISO/IEC 25010 to assess system quality across usability, reliability, and security dimensions [9], [10]. Additional metrics such as intent recognition accuracy, response latency, and conversation length have been used to quantify system effectiveness and engagement [8].

   6. Research Gap

   Although existing studies validate the effectiveness of chatbots in mental health and education, limited work has focused on institution-based, expert-guided chatbot systems deployed within Philippine higher education settings. LumiCHAT addresss this gap by integrating expert- validated mental health guidance, privacy-conscious system design, and usability evaluation tailored to student support services.

3. METHODOLOGY

   This study adopted the Agile Software Development Life Cycle (SDLC) model for the development of LUMICHAT, an expert-based chatbot for mental health support for students. Agile was selected for its iterative, collaborative, and user-centered nature, which enables continuous refinement based on stakeholder feedback and evolving user needs. The Agile cycle consists of planning, design, development, testing, deployment, and review phases, supporting responsiveness to change throughout the system lifecycle.

   Figure 1.0 Agile lifecycle, adapted from Pratama & Kristiana

   (2023)

   The development of LUMICHAT involved active participation from guidance counselors, mental health professionals, and student users. Agile facilitated a participatory environment where functional components could be incrementally delivered, evaluated, and refined to ensure alignment with both technical and psychological requirements. This approach ensured that system accuracy and emotional appropriateness were addressed consistently across development phases [16].

   1. Planning Phase

      During the planning phase, the research team conducted structured interviews and consultations with guidance counselors and students to identify functional and emotional requirements for LUMICHAT. These activities aimed to surface challenges related to access to mental health support, emotional expression, anonymity, and availability.

      This stakeholder-driven approach aligns with prior chatbot development studies in mental health contexts, where end-user participation significantly influenced system design and feature prioritization. Through co-design, key features such as anonymous interaction, mood tracking, and counselor referral booking were identified [10].

      User requirements were translated into structured user stories following the format: As a [user role], I want [feature] so that [benefit]. These user stories formed the initial product backlog and guided sprint planning.

      To contextualize existing limitations, a manual process flowchart was created to document current counseling workflows and pain points, including long waiting times and limited counselor availability. Process visualization has been shown to effectively reveal inefficiencies and guide digital intervention design [15].

   2. Design Phase

      The design phase focused on translating user stories into tangible system representations. Low- and high-fidelity wireframes were developed using Figma to visualize the user interface and interaction flow. The designs followed inclusive UX principles to ensure accessibility and emotional comfort across devices.

      The interface design was informed by affective computing and emotionally aware UX guidelines, emphasizing clarity, reassurance, and trust in userchatbot interactions [8].

      Use-case diagrams were created to model interaction scenarios such as initiating conversations, expressing emotional distress, receiving coping suggestions, and escalating to counselor support. These artifacts helped stakeholders validate functional coverage.

      The emotional tone and phrasing of chatbot responses were reviewed based on best practices in empathetic conversational design, ensuring that responses acknowledged user emotions and avoided generic or dismissive language [13], [17].

   3. Development & Implementation (Tools and Software)

      The development phase implemented LUMICHAT through iterative Agile sprints emphasizing modularity, collaboration, and continuous integration. The system was designed using a layered architecture that separated presentation, application logic, chatbot processing, and data management to improve maintainability and scalability [9].

      The LumiCHAT system is organized as a four-layer architecture (see Fig. 2.0). The presentation layer (React.js with Tailwind/Bootstrap) implements the responsive user interface chat window, mood indicators, response cards, and counselor prompts and enforces client-side accessibility and validation [8]. The application layer (Laravel) handles authentication, session management, business logic, and REST APIs that mediate client requests and backend services. The chatbot engine layer (Rasa) hosts the natural language understanding and dialogue management pipelines responsible for intent detection, policy-driven dialogue control, fallback handling, and response selection. (MySQL + artifact storage) stores pseudonymized logs, conversation histories, model artifacts, and administrative records while supporting secure retrieval and analytics for monitoring and model refinement.

      During development the stack evolved from a locally hosted prototype to a live deployment: the Laravel application was hosted on a secure server, while the Rasa engine ran on a dedicated VPS and communicated with the application over authenticated HTTP endpoints; version control (GitHub) and CI practices ensured traceability and code quality.

      Training data pipelines included text collection, cleaning, anonymization, counselor-validated labeling, conversion to Rasa YAML artifacts (nlu.yml, domain.yml, rules.yml), iterative training, and evaluation (precision / recall / F1 and confusion analysis) prior to production packaging. These processes supported progressive improvements in intent recognition and response relevance. [10] [11] [12]

      Figure 3.0 RASA NLU and Data Lifecycle Pipeline for LumiCHAT

      The Rasa pipeline includes data preparation, YAML conversion, pipeline configuration (config.yml), training, evaluation, and packaging; datasets and anonymization steps were applied to training data [11].

   4. Testing Phase (planned measures)

      Testing combined functional verification of end-to-end workflows and usability evaluation using the System Usability Scale (SUS). Planned evaluation metrics included SUS scores, intent-classification performance (precision, recall, F1-score), referral precision, and task-completion

      times [7]. Detailed test artifacts are provided in Figure 5.0 , Figure 6.0 and Table 1.0.

   5. Deployment and Review

   LUMICHAT was deployed to a secure live hosting environment to ensure accessibility beyond school hours. All interactions were anonymized and encrypted to protect user privacy. Deployment was supported by onboarding sessions, user guides, and in-app prompts.

   The review phase involved collecting qualitative feedback from students and counselors through interviews and surveys. Feedback and system logs informed iterative improvements during sprint retrospectives, focusing on response tone, clarity, and escalation accuracy.

4. RESULTS & DISCUSSION

   1. Implementation Outcomes

      LUMICHAT was implemented with the core features defined during the design and development phases, including an NLU-driven conversational engine, anonymous interaction mode, appointment booking, and counselor administration tools. The deployed system was functionally evaluated through scenario-based testing conducted with student participants and guidance counselors to verify end- to-end workflow completion.

      Figure 4.0 High Fidelity Web Interface for Chatbot Interface in LumiCHAT

      Figure 4 shows the implemented student chat interface used during scenario-based testing. The interface supports core user tasks such as initiating conversations, receiving system responses, and accessing support features. Test observations indicate stable task completion with minimal operational issues, while user feedback highlighted areas for improvement in navigation clarity and the specificity of coping guidance.

   2. NLP & Model Performance

      The chatbos natural language understanding model was evaluated using standard intent-classification metrics derived from a held-out test dataset. Performance results demonstrated acceptable precision, recall, and F1-scores across most defined intents, with stronger performance observed for clearly delineated categories (e.g., burnout and

      time management) and comparatively lower performance for semantically overlapping emotional intents.

      Figure 5.0 summarizes the confusion matrix and intent- level performance metrics, highlighting instances where semantic overlap between emotional intents contributed to misclassification. These findings informed iterative refinements to training data and dialogue rules in subsequent development cycles.

      Figure 6.0 the response selection results indicate consistent ranking accuracy for predefined response candidates, supporting reliable system behavior during conversational exchanges.

   3. Usability & Task Performance

      System usability was assessed using the System Usability Scale (SUS) following completion of predefined interaction tasks. Participants representing different user roles completed the SUS questionnaire after hands-on use of the system.

      Table 1.0 Final SUS Score Computation for LumiCHAT

      The aggregated SUS score indicated overall usability within the acceptable range. Analysis of task performance and participant feedback identified specific usability issues related to interface navigation and the clarity of suggested coping actions. These issues were addressed through iterative interface adjustments and response template refinement, demonstrating the effectiveness of the Agile feedback-driven development process.

   4. Discussion (concise)

      • The layered architecture with an on-premises Rasa deployment balanced privacy and institutional control while delivering acceptable NLU performance; improving detection of nuanced emotional intents will require larger and more diverse labeled datasets.

      • Usability scores indicate the system is ready for limited roll-out while content personalization and

   longitudinal outcome tracking are left for future work.

5. Conclusion & Future Work

LUMICHAT is a practical, privacy-aware chatbot that complements campus counseling by providing early- intervention conversational support and an automated referral pathway to licensed counselors. The implemented system demonstrated acceptable usability and intent- classification performance sufficient for initial risk triage and limited roll-out.

Future improvements include (1) expanding and diversifying labeled datasets to improve detection of nuanced emotional intents, (2) adding personalization features while preserving user privacy, (3) integrating speech and other multimodal inputs, (4) implementing a continuous monitoring and model-update pipeline, and (5) conducting a longitudinal outcomes study to measure changes in help-seeking behavior and clinical impact.

ACKNOWLEDGMENT

The authors gratefully acknowledge the guidance and supervision of Mr. Jan James M. Pada, whose advice significantly improved the project direction and manuscript. We also thank Mr. Jason Ang for his stakeholder contributions and for facilitating participant recruitment. Our sincere appreciation goes to the counselors and students of Tagoloan Community College who participated in the needs assessment and completed the System Usability Scale (SUS) evaluations; their time and candid feedback were essential to the systems refinement. We are likewise thankful for the technical contributions of the development team and the support provided by project supervisors during design and implementation. Correspondence concerning this article should be addressed to lumichatproject@gmail.com.

REFERENCES

1. World Health Organization. (2022). World mental health report: Transforming mental health for all. Geneva: World Health Organization. https://creativecommons.org/licenses/by-nc-sa/3.0/igo/

2. UNICEF. (2021). COVID-19 and school closures: One year of education disruption. The United Nations. Retrieved March 15, 2025, from https://coilink.org/20.500.12592/7h44mgg

3. Pedrelli, P., Nyer, M., Yeung, A., Zulauf, C., & Wilens, T. (2014). College students: Mental health problems and treatment considerations. Academic Psychiatry, 39(5), 503.

   https://doi.org/10.1007/s40596-014-0205-9

4. Lee, J. Y. (2023). Can an artificial intelligence chatbot be the author of a scholarly article? Journal of Educational Evaluation for Health Professions, 20(6). https://doi.org/10.3352/jeehp.2023.20.6

5. van der Schyff, E. L., Ridout, B., Amon, K. L., Forsyth, R., & Campbell, A. J. (2023). Providing self-led mental health support through an artificial intelligencepowered chatbot (Leora) to meet the demand of mental health care. Journal of Medical Internet Research, 25, e46448. https://doi.org/10.2196/46448

6. Hasal, M., Nowaková, J., Saghair, K. A., Abdulla, H., Snáel, V., & Ogiela, L. (2021). Chatbots: Security, privacy, data protection, and social aspects. Concurrency and Computation: Practice and Experience, 33(19), e6426. https://doi.org/10.1002/cpe.6426

7. Manole, A., Cârciumaru, R., Brînzas, R., & Manole, F. (2025). An exploratory investigation of chatbot applications in anxiety

   management: A focus on personalized interventions. Information, 16(1), 11. https://doi.org/10.3390/info16010011

8. Skurdin, A., & Wynn, M. (2024). Agile methodologies in chatbot development: Bridging strategic goals and development practices. International Journal of Software Engineering and Knowledge Engineering, 34(1), 2540. https://www.mdpi.com/2078- 2489/15/4/226

9. Alegado, R. T., Luciano, R. G., Alcantara, G. M., Laureta, A. L., Abando, B. C. Y., & Makidang, J. P. S. (2022). Design and development of chatbot request system. International Journal of Science and Applied Information Technology, 11(3), 15. https://doi.org/10.30534/ijsait/2022/011132022

10. Singla, A., Khanna, R., Kaur, M., Kelm, K., Zaiane, O., Rosenfelt, C. S., & Majnemer, A. (2024). Developing a chatbot to support individuals with neurodevelopmental disorders: Tutorial. Journal of Medical Internet Research, 26, e50182. https://doi.org/10.2196/50182

11. Pérez-Soler, S., Juarez-Puerta, S., Guerra, E., & de Lara, J. (2021). Choosing a chatbot development tool. IEEE Software, 38(3), 7279. https://doi.org/10.1109/MS.2020.3030198

12. Abdellatif, A., Shihab, E., & Hassan, A. E. (2020). Challenges in chatbot development: A study of Stack Overflow posts. In Proceedings of the 17th International Conference on Mining Software Repositories (pp. 275285). Association for Computing Machinery. https://doi.org/10.1145/3379597.3387498

13. Zhou, L., Gao, J., Li, D., & Shum, H.-Y. (2020). The design and implementation of XiaoIce, an empathetic social chatbot. Computational Linguistics, 46(1), 5393. https://doi.org/10.1162/coli_a_00368

14. Kingchang, T., Nguyen, H. T., & Lee, S. (2024). AI-powered chatbots in academic and career decision-making: A study on student trust and alignment. Journal of Educational Technology & Society, 27(2), 112

    125. https://www.ijiet.org/vol14/IJIET-V14N1-2021.pdf

15. Cuevas, J. M., Dela Cruz, M. A., & Santos, L. R. (2024). Development of an NLP-integrated chatbot for barangay-level public service systems in the Philippines. Philippine Journal of Information Systems, 10(1), 4560. http://pdfs.semanticscholar.org/0c04/909e00c4bfab5994b8f6fa26a8d 3fc19e116.pdf

16. Pratama, D. I., & Kristiana, A. S. (2023). Applying Scrum methodology in software development: Lessons from academic projects. Journal of Software Engineering Practices, 12(1), 1529. https://pdfs.semanticscholar.org/f7d5/31878fef7aa7c1195ea23b2c4f9 11d23a2e4.pdf

17. Macaraeg, J. E., & Lacsamana, A. T. (2024). Enhancing conversational design for empathetic interactions in mental health chatbots. Journal of AI and Society, 19(3), 321335. Pérez-Soler, S., Juarez-Puerta, S., Guerra, E., & de Lara, J. (2021). Choosing a chatbot development tool. IEEE Software, 38(3), 7279. https://doi.org/10.1109/MS.2020.3030198

18. Abd-Alrazaq, A. A., Alajlani, M., Ali, N., Denecke, K., Bewick, B. M., & Househ, M. (2021). Perceptions and opinions of patients about mental health chatbots: Scoping review. Journal of Medical Internet Research, 23(1), e17828. https://doi.org/10.2196/17828

19. Arjona, B. R., Crisologo Jr, D., Dela Rosa, M. I., Montilla, M. G., Narvaez, R. A., & Suarez, R. F. (2023). Impact of digital health apps among patients with mental health issues: An integrative review. Canadian Journal of Nursing Informatics, 18(1). https://cjni.net/journal/?p=10879

20. Bantjes, J., Kessler, M. J., Hunt, X., et al. (2023). Treatment rates and barriers to mental health service utilization among university students in South Africa. International Journal of Mental Health Systems, 17,

    38. https://doi.org/10.1186/s13033-023-00605-7

21. Campbell, F., Blank, L., Cantrell, A., et al. (2022). Factors that influence mental health of university and college students in the UK: A systematic review. BMC Public Health, 22, 1778. https://doi.org/10.1186/s12889-022-13943-x

22. Maravilla, N. M., & Tan, M. J. (2021). Philippine Mental Health Act: Just an Act? A call to look into the bi-directionality of mental health and economy. Frontiers in Psychology, 12, 706483. https://doi.org/10.3389/fpsyg.2021.706483

23. Sifat, M., Huq, M., Baig, M., Tasnim, N., & Green, K. M. (2023). An examination of barriers to accessing mental health care, and their association with depression, stress, suicidal ideation, and wellness in a

    Bangladeshi university student sample. International Journal of Environmental Research and Public Health, 20(2), 904. https://doi.org/10.3390/ijerpp0020904

24. Taylor, K. V., Garchitorena, L., Scaramutti-Gladfelter, C., Wyrick, M., Grill, K. B., & Seixas, A. A. (2024). A digital mental health solution to improve social, emotional, and learning skills for youth: Protocol for an efficacy and usability study. JMIR Research Protocols, 13, e59372. https://doi.org/10.2196/59372

25. Wijekoon Mudiyanselage, K. W., De Santis, K. K., Jörg, F., Saleem, M., Stewart, R., Zeeb, H., & Busse, H. (2024). The effectiveness of mental health interventions involving non-specialists and digital technology in low-and middle-income countries A systematic review. BMC Public Health, 24, 77. https://doi.org/10.1186/s12889- 023-17417-6

APPENDICES

‌Figure 2.0 Layered Architecture for LumiCHAT

Figure 5.0 Model Evaluation (Intent Confusion Matrix)

Figure 6.0 Model Evaluation (Response Selection Metrics)