Abstract
Social recommendation aims to fuse social links with user-iteminteractions to alleviate the cold-start problem for rating prediction.Recent developments of Graph Neural Networks (GNNs) motivateendeavors to design GNN-based social recommendation frame-works to aggregate both social and user-item interaction informa-tion simultaneously. However, most existing methods neglect thesocial inconsistency problem, which intuitively suggests that so-cial links are not necessarily consistent with the rating predictionprocess. Social inconsistency can be observed from both context-level and relation-level. Therefore, we intend to empower the GNNmodel with the ability to tackle the social inconsistency problem.We propose to sample consistent neighbors by relating samplingprobability with consistency scores between neighbors. Besides, weemploy the relation attention mechanism to assign consistent rela-tions with high importance factors for aggregation. Experimentson two real-world datasets verify the model effectiveness.
Liangwei Yang
Research Scientist at Salesforce Research
My research interests include graph, recommender system and efficient modeling.