Self-Learning Chatbots using Reinforcement Learning
Mohsin Bashir Lone, Nahida Nazir, Navneet Kaur, Dinil Pradeep, Ammad Ul Ashraf, Pir Asrar Ul Haq, Nouman Barkat Dar, Abid Sarwar, Manik Rakhra, Omdev Dahiya · 2022 3rd International Conference on Intelligent Engineering and Management (ICIEM) · 2022
There has been a surge in the development of dialogue generation systems also called chatbots in Recent years. Research on sequence-to-sequence architectural chatbots has led to a number of efficient and adaptive chatbots. Several of them worked on goal-oriented chatbots, some worked on open domain chatbots and others developed emotional chatbots called CheerBots that respond to people’s emotions. Be it in the IT field or in Education, these Dialogue systems have proven their worth in every industry. This paper discusses, the proposal of a SEQ2SEQ architecture-based chatbot that uses a Reinforcement Learning Algorithm to respond to user queries. This allows the model to explore the domain of all possible responses that can be generated by the SEQ2SEQ model by learning novel utterances/queries. The model answered all the queries, making it a suitable chitchat system for open-domain applications like SIRI and ALEXA. It responded in an engaging and interactive manner which led to an interesting conversation. The model, despite being complex, showed remarkable accuracy.