Mental Health Chatbot Using Deep Learning and Natural Language Processing

Abstract

In the present digital age, the exponential growth of mental health assistance online has made information extraction crucial and challenging. To address this, we developed an AI mental health chatbot using Neural Networks and Natural Language Processing (NLP) to provide concise and understandable responses to users seeking support. Mental health support dataset was sourced from Kaggle and preprocessed it using tokenization, text cleaning, and named entity recognition, generating feature vectors that represent the contextual relevance of phrases. Additionally, a Recurrent Neural Network (RNN) was utilized to generate contextually appropriate responses. This study employed Long Short-Term Memory (LSTM) and Bi-directional Long Short-Term Memory (Bi-LSTM) to extract mental health keywords. Bi-LSTM achieved a lower validation loss (0.4852) than LSTM (0.5321), but LSTM was selected for the model due to superior performance in response generation. Notably, Bi-LSTM's validation loss increased significantly (1.8078) when generating responses, indicating its limitations in capturing contextual dependencies. We attribute LSTM's suitability to its ability to preserve temporal relationships and contextual information, crucial for generating empathetic and personalized responses. In contrast, Bi-LSTM's increased complexity and bidirectional processing may have introduced unnecessary dependencies, hindering response generation. The findings of this study have highlighted the robustness and interpretability of the chatbot's responses, the importance of computational resources, and the necessity for domain-specific knowledge in selecting the most suitable approach for providing effective mental health support through text-based interactions.