AntRS: Recommending Lists Through a Multi-objective Ant Colony System

摘要

When people use recommender systems, they generally expect coherent lists of items. Depending on the application domain, it can be a playlist of songs they are likely to enjoy in their favorite online music service, a set of educational resources to acquire new competencies through an intelligent tutoring system, or a sequence of exhibits to discover from an adaptive mobile museum guide. To make these lists coherent from the users' perspective, recommendations must find the best compromise between multiple objectives (best possible precision, need for diversity and novelty). We propose to achieve that goal through a multi-agent recommender system, called AntRS. We evaluated our approach with a music dataset with about 500 users and more than 13,000 sessions. The experiments show that we obtain good results as regards to precision, novelty and coverage in comparison with typical state-of-the-art single and multi-objective algorithms.