AI-Powered Mental Health Chatbot: A Research Paper

Abstract

Mental health disorders, including anxiety and depression, affect millions worldwide, yet many individuals refrain from seeking professional help due to stigma, accessibility challenges, and financialconstraints.ThispaperpresentsanAI-poweredmentalhealthchatbotdesignedtoprovide empathetic, accessible, and anonymous emotional support using Generative AI and Natural Language Processing (NLP). The chatbot offers 24/7 availability, confidential conversations, and self-care recommendations while ensuring data privacy and security. Through comparison with existingsolutions, thisstudyhighlightsits superiorcontextualawarenessandengagementfeatures. Future enhancements, including voice-based interaction and multilingual support, are explored.

Introduction

1. Introduction

Mental health disorders are a growing global concern, but many individuals face barriers like stigma, high costs, and lack of access to traditional therapy. AI-driven chatbots offer a promising alternative by providing instant, private, and free emotional support.

2. Problem Statement

Despite available services, many people do not seek help due to accessibility issues and social stigma. This project proposes a 24/7 AI-powered chatbot that offers confidential emotional support and self-care guidance.

3. Objectives

• Deliver empathetic, real-time conversations using NLP and sentiment analysis.

• Ensure user privacy and anonymity.

• Provide personalized self-care techniques and interactive UI.

• Increase user engagement and comfort through design and accessibility.

4. Literature Review

• AI Evolution: Chatbots like Woebot, Wysa, and Replika show the usefulness of AI in mental health.

• Model Advancements: The chatbot uses Gemma-2-2b-it, a Transformer-based LLM, fine-tuned on mental health datasets for emotionally intelligent responses.

• Ethical Concerns: Important to address data privacy, accuracy, and the need for disclaimers recommending professional help.

5. Methodology

• System Design:

  • UI built with Streamlit for interactive user experience.

  • Backend handled via FastAPI for secure AI interactions.

  • AI processing through Gemma-2-2b-it, trained on relevant datasets.

• Model Training:

  • Used mental health data from HuggingFace.

  • Applied tokenization, intent recognition, and sentiment analysis.

  • Evaluated using BLEU score, perplexity, and human feedback.

6. Results & Discussion

• Model Performance:

  • BLEU score > 0.6 and perplexity ~10–15, indicating relevant and coherent responses.

• User Feedback:

  • Among 50 test users, 85% reported satisfaction, citing emotional resonance and engagement.

• Future Enhancements:

  1. Voice-based interaction (speech-to-text).

  2. Multilingual support for wider reach.

  3. Hybrid AI-therapist model to combine automation with human guidance.

Conclusion

Thisresearchhighlightsthepotential ofAI-poweredchatbotsinmentalhealthsupport,addressing stigma, accessibility, andaffordability challenges. While AI cannot replacetherapists, itserves as an effective first-line support system. Future advancements, including voice interaction and sentiment-based AI recommendations, will further enhance user engagement and effectiveness.

References

[1] Bickmore, T. W., & Picard, R. W. (2005). Establishing and Maintaining LongTerm Human- Computer Relationships. ACM Transactions on Computer-Human Interaction, 12(2), 293–327. [2] Fitzpatrick, K. K., Darcy, A.,& Vierhile, M. (2017). DeliveringCognitiveBehavior Therapy to Young Adults With Symptoms of Depression and Anxiety Using a Fully Automated ConversationalAgent(Woebot):ARandomizedControlledTrial.JMIRMentalHealth,4(2),e19. [3] Luxton, D. D.,June, J. D.,&Sano, A. (2019). ArtificialIntelligenceinBehavioralandMental Health Care. Academic Press. [4] Miner,A.S.,Milstein,A.,Hancock,J.T.,&etal.(2016).Smartphone-BasedConversational Agents and Responses to Questions About Mental Health, Interpersonal Violence, and Physical Health. JAMA Internal Medicine, 176 [5] 619–625.[5]GoogleAI.(2024).Gemma2Model-ALightweightOptimizedLLMfor Conversational AI. Retrieved from https://huggingface.co/google/gemma-2-2b-it [6] MetaAI.(2022).BlenderBot:ConversationalAIModelforOpen-DomainDialogue. [7] Retrievedfromhttps://huggingface.co/facebook/blenderbot-small [8] Hugging Face. (2023). Mental Health Dataset \"ramachaitanya22/mental_health_and_fitness_data\". Retrieved - from https://huggingface.co/datasets/ramachaitanya22/mental_health_and_fitness_data [9] WorldHealthOrganization. (2021). MentalHealthandCOVID-19:EarlyEvidenceofthe Pandemic’s Impact. WHO Press. [10] OpenAI. (2023).TheEvolutionofAIChatbotsinMentalHealthSupport.Retrievedfrom https://openai.com/research [11] Streamlit. (2024). Building Interactive Web Apps with Streamlit. Retrieved from https://docs.streamlit.ioCOET,AKOLA/B.E/CSE/2024-202534AI-PoweredMentalHealth Chatbot . [12] HuggingFaceDatasets.(2023).\"ramachaitanya22/mental_health_and_fitness_data\". Retrived from- https://huggingface.co/datasets /ramachaitanya22/mental_ health_and_fitness_data

Copyright

Copyright © 2025 Prof. S. S. Dixit, Shruti Mahulkar, Manasi Pandao, Gauri Rathor, Janhavi Thokane. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.