An interactive user groups recommender system based on reinforcement learning

Nowadays, we have access to countless and diverse user data in various fields. Thus, it requires analysis to find a set of users. Several steps are taken to understand and identify users interactively to achieve such a goal. This article introduces a reinforcement learning model of interactive recommendations based on user groups. An agent learns the appropriate policy to discover users among groups based on feedback during a sequential decision-making process and recommends the best action for the next step. There are three datasets available for courses, jobs, and LinkedIn, but these three datasets are not related to each other, which causes errors in learning politics. Furthermore, taking different actions will significantly affect learning from these datasets. To improve learning, semantic similarity and text processing are used to extract relationships between datasets. As a result, a set of groups is constructed based on what the users have in common. Users are chosen from a set of groups represented by an agent. The results and experiments show that the agent can learn the policy without collecting previous sessions and finally provide an acceptable recommendation.