SVD-based group recommendation approaches: An experimental study of Moviepilot

Nowadays most recommender systems are made for individuals. However, there is a need to offer recommendations to a group rather than an individual in many scenarios, such as interactive TV watched in a family, friends traveling together. Taking consideration of group information as an additional context has also been a challenge in context-aware recommender systems. In this work, we propose some SVD-based group recommendation methods through aggregating ratings of group members with different group decision strategies, including weighted, least misery, and hybrid ones. These methods are divided into two categories: SVD-based aggregation profiles and aggregation predictions methods. The former ones employ "group aggregation first, SVD-based prediction later", while the latter ones are opposite. Finally we conduct some experiments on the Moviepilot dataset released for the Challenge on Context-Aware Movie Recommendation (CAMRa2011) to evaluate the effectiveness of different SVD-based group recommendation approaches, and analyze the results.