Non-pairwise Collaborative Filtering

On the users’ interaction graph, neighbors have been widely explored in the embedding function of collaborative filtering to address the sparsity issue. However, the embedding learning models are highly subject to the following pairwise interaction function on interest prediction. We argue that the core of personalized recommendation locates interaction rather than embeddings. Distinct from the sparse pairwise interactions, there are a large amount of inherent non-pairwise signals hidden among neighbors, which are promising for interaction learning. In this work, we explore the active effect of non-pairwise neighbors on the target user-item pair and propose a non-pairwise collaborative filtering (NPCF) model. For a target user-item pair, NPCF mines target-aware CF signals of neighbors by aggregating both pairwise and non-pairwise CF signals of the target for a target-specific interaction embedding. Experiments on three real-world datasets demonstrate that NPCF outperforms the state-of-the-art models for personalized recommendation. It implies NPCF is capable of learning interactions with non-pairwise neighbors.