Exploration in Recommender Systems

Recommender systems have experienced increasing success in delivering personalized content recommendations to users. However, these systems often rely heavily on popular content, neglecting the need to capture the continuously evolving interests of users. There is no direct method to systematically explore and understand users’ interests, which has a negative impact on the overall quality of the recommendation process. The training data for these systems is generated from the content candidates presented to the user, further exacerbating the issue. In this talk, we will dive deeper into the various types of exploration paradigm being used in recommendation systems. We will provide a comprehensive overview of existing exploration approaches, thoroughly examining their strengths and limitations. The key challenge in this field of research lies not only in finding efficient exploration strategies but also in devising effective measurements to assess their effectiveness. By shedding light on these intricacies, our aim is to inspire further work into understanding and enhancing exploration in recommender systems.