Modelling High-Order Social Relations for Item Recommendation

The prevalence of online social network makes it compulsory to study how relations affect user choice. However, most existing methods leverage only first-order relations, that is, the direct neighbors are connected target user. high-order e.g., friends friends, which very informative reveal preference, have been largely ignored. In this work, we focus on modeling indirect influence from in networks improve performance item recommendation. Distinct mainstream recommenders regularize model learning with instead propose directly factor predictive model, aiming at better embeddings To address challenge increase dramatically order size, recursively “propagate” along network, effectively injecting into representation. We conduct experiments two real datasets Yelp and Douban verify our High-Order Social Recommender (HOSR) model. Empirical results show HOSR significantly outperforms recent graph regularization-based NSCR IF-BPR $^+$ , convolutional network-based prediction DeepInf, achieving new state-of-the-arts task.