Integrating sentiment analysis to enhance mental health support chatbots

Mental health conditions have adverse effects on an individual's quality of life. Between 2017 and 2019, there were 52.9 emergency department visits per 1,000 adults for mental health disorders. The rising popularity of artificial intelligence prompted its use in depression detection or mental condition diagnosis, establishing its prominent role in the mental health sector. In this research, we developed an AI chatbot with enhanced contextual intelligence using sentiment analysis to support individuals experiencing mental distress. For this research, we utilized a dataset of text data from 1.6 million posts on the social media platform X (formerly Twitter) due to its open-source availability and accessibility. The collected dataset was cleaned of repetitive characters, special characters, URLs, and numbers. NLP techniques, including tokenization, lemmatization, and stemming, were used for pre-processing. The application identifies sentiments, generates responses, and suggests basic support remedies. We used the Rasa framework to create a hybrid chatbot with customizable configurations. An LSTM model for sentiment analysis was integrated into the Rasa chatbot as a custom action component. A batch size of 32 and an optimal maximum sequence length were selected for balanced training efficiency and accuracy. The LSTM model achieved 76% accuracy in training and validation, enhancing the chatbot text comprehension. Future improvements will include adding personal features and expanding the user base.