A stochastic approach to group recommendations in social media systems.

Recommender systems are becoming increasingly important not only to individual users but also to groups of people. This study focuses on the issue of recommending items to communities of interest (i.e., groups) that are specifically formed in social media systems. To deal with this issue, we introduce a new graph model that profits from fruitful tagging information. By using the proposed graph model, we present a stochastic method that makes recommendations based on link-structure analysis in a probabilistic manner. This method supports two ways of computing group ranking scores for items—via a preference aggregation approach and via a ranking aggregation approach, but ensures the same ranking results. We also explore the influence of users and items associated with a group in the facilitation of more accurate recommendations. Our empirical evaluations with the Last.fm dataset corroborate the benefits of our graph model on group recommendations, and demonstrate that the proposed group recommendation method performs better than existing alternatives.