Retrieval based Emotionally responsive chatbot Utilizing emotions clustering and XAI Techniques.

Abstract

Chatbots are becoming increasingly prevalent in our daily lives, handling a variety of tasks from booking flights to checking the balance of a bank account. However, these interactions are often limited to transactional nature, lacking the depth and nuance of natural human conversations. Therefore, while chatbots are becoming more sophisticated, they still have a significant way to go in terms of authentic human-like interaction. Enhancing chatbots' ability to interact with humans and making them more user-friendly is a crucial challenge in contemporary conversational technology. One promising approach is to imbue these chatbots with the ability to express and understand emotions. Hence, the focus is on developing empathetic dialogue systems that effectively model empathy, ultimately making chatbots more human-like in their interactions with users. In this project, A retrieval-based approach is adopted to mitigate the limitations encountered in generative-based models, particularly concerning logical coherence and response appropriateness. Unlike generative models, retrieval-based approaches offer a more pragmatic solution that demands fewer computational resources. Through the integration of deep learning techniques, transparency in decision-making processes is ensured by eXplainable Artificial Intelligence (XAI), providing users with insights into the chatbot's reasoning. Furthermore, clustering techniques are leveraged to generalize various emotion types, allowing the chatbot to cater to a broader spectrum of emotional inputs effectively.