Exploring Response Functions for Recommender Systems in Simulated Environment

Recommender systems have become an essential feature of online platforms, assisting users in navigating the large quantity of information and options accessible. However, creating and assessing these systems is a challenging task since it requires understanding and modeling of complicated user behavior. Simulators, which model user interactions within recommender systems, provide a versatile and realistic method for evaluating these systems. One of the key components of such simulators is the response function, which models how users respond to recommendations. This study investigates the efficacy and significance of various response functions within recommender system simulators. I present comparison of machine learning models for modeling response functions, which take into account features that relate to items, users and their interactions. I evaluate the approaches using both a classical classification approach, and a simulated environment, comparing them to each other and a baseline in terms of accuracy and predictive power. This paper contributes to the growing body of work on response functions in recommender system simulators, offering valuable insights and directions for future research.