Social affinity filtering: recommendation through fine-grained analysis of user interactions and activities

Content recommendation in social networks poses the complex problem of learning user preferences from a rich and complex set of interactions (e.g., likes, comments and tags for posts, photos and videos) and activities (e.g., favourites, group memberships, interests). While many social collaborative filtering approaches learn from aggregate statistics over this social information, we show that only a small subset of user interactions and activities are actually useful for social recommendation, hence learning which of these are most informative is of critical importance. To this end, we define a novel social collaborative filtering approach termed social affinity filtering (SAF). On a preference dataset of Facebook users and their interactions with 37,000+ friends collected over a four month period, SAF learns which fine-grained interactions and activities are informative and outperforms state-of-the-art (social) collaborative filtering methods by over 6% in prediction accuracy; SAF also exhibits strong cold-start performance. In addition, we analyse various aspects of fine-grained social features and show (among many insights) that interactions on video content are more informative than other modalities (e.g., photos), the most informative activity groups tend to have small memberships, and features corresponding to ``long-tailed'' content (e.g., music and books) can be much more predictive than those with fewer choices (e.g., interests and sports). In summary, this work demonstrates the substantial predictive power of fine-grained social features and the novel method of SAF to leverage them for state-of-the-art social recommendation.