Session Based Recommender System: A Generative Approach

In the age of information overload and digitalization, recommender systems (RSs) have become increasingly important for informed decision-making and service utilization. Session-based recommender systems (SBRSs) have emerged as a new type of RS in recent years. Unlike traditional RSs like content-based RSs and collaborative filtering-based RSs, which focus on static, long-term user preferences, SBRSs aim to capture dynamic, short-term user preferences to provide timely and accurate recommendations that adapt to evolving user contexts. This research propose novel generative model that capture the patterns among the user's choices and generate the items for the given session. We observed that pattern recognition algorithms only capture past user choices or the choices of similar users. To address this limitation, we propose a non-deterministic predictive model that balances exploration and exploitation.