Evaluating recommender system using multiagent-based simulator: Case study of collaborative filtering simulation

Abstract

This paper describes a agent-based simulation system to evaluate recommender systems. Recommender systems have attracted attention to present items found by preference of users. Many algorithms for recommender system are developed but the comparisons between their algorithms are difficult because of limited data set and the difficulty of constructing simulator environment. In order to resolve them, we develop agent-based recommender system simulator. This simulator constructs the simulator environment based on the network model, and lets recommender agent recommend items to agents, evaluates the items, and summarizes(outputs) the recommendation results. In the experiment on 100 agents, we can confirm the usability of our simulator because of recapturing the feature of collaborative filtering by this simulator. © 2013 Springer-Verlag GmbH.