Personalized Chatbot Responses using Reinforcement Learning and User Modeling

Abstract

The research focuses on chatbot interaction enrichment using reinforcement learning and user modeling. It aims to develop a personalized RL-based response generation framework for the optimization of satisfaction, engagement, and completion rates for the users. Anonymized historical interaction data from 500 users were collected to generalize user-profiles and contextual models. This was done by extracting features and observing and evaluating the control and experimental groups to test the efficacy of the working personalized system. Results indicated an overall increase of 35 percent in user satisfaction, 50 percent in session duration, and 25 percent in rates of completion of tasks against the traditional rule-based system. The results are very much in line with current literature on the improvements personalization can bring to the user's experience across many domains. The results from this study thus propose that personal AI systems powered with fine-grained models of users and reinforcement learning could obtain more engaging and efficient user interactions. The result has further-reaching implications for, for example, e-learning, customer service, and healthcare applications.