Group recommendation with temporal affinities

We examine the problem of recommending items to ad-hoc user groups. Group recommendation in collaborative rating datasets has received increased attention recently and has raised novel challenges. Different consensus functions that aggregate the ratings of group members with varying semantics ranging from least misery to pairwise disagreement, have been studied. In this paper, we explore a new dimension when computing group recommendations, that is, affinity between group members and its evolution over time. We extend existing group recommendation semantics to include temporal affinity in recommendations and design GRECA, an efficient algorithm that produces temporal affinity-aware recommendations for ad-hoc groups. We run extensive experiments that show substantial improvements in group recommendation quality when accounting for affinity while maintaining very good performance.