Emotion Driven Chatbot Using Natural Language Generation Techniques

The increasing developments in chatbot architectures introduce a choice between lack of generality in the chatbot responses or an increase in the complexity of optimization. To address this compromise, we propose a divide and conquer approach in the architecture of the chatbot. This behavior in the chatbot is demonstrated by making the chatbot emotionally and contextually aware of the user's intent while maintaining a knowledge base for world knowledge-based queries. We propose a chatbot system which can generate more natural responses based on the abstract emotional knowledge acquired by the user input and the database of pre-labeled responses. The responses are generated using sequence to sequence model introduced in Learning Phase Representation using RNN Encoder- Decoder for statistical Machine Translation. The goal is to generate responses which can be helpful in changing users’ emotional state while satisfying some criteria to measure the effectiveness of the system. Chatbot here will provide the basic conversation functionality based on emotional state of user and also carry on a normal day to day conversation. Several chatbots are available but they require user to setup key phrases manually and does not work for different set of key phrases other than the knowledge base. Our proposed system ranks every response, every time, they interact with a user and also lead to addition of new phrases and responses in sequential interactions with the user.