Multi-Behavior Recommendation with Cascading Graph Convolution Networks

Abstract

Multi-behavior recommendation, which exploits auxiliary behaviors (e.g.,\nclick and cart) to help predict users' potential interactions on the target\nbehavior (e.g., buy), is regarded as an effective way to alleviate the data\nsparsity or cold-start issues in recommendation. Multi-behaviors are often\ntaken in certain orders in real-world applications (e.g., click>cart>buy). In a\nbehavior chain, a latter behavior usually exhibits a stronger signal of user\npreference than the former one does. Most existing multi-behavior models fail\nto capture such dependencies in a behavior chain for embedding learning. In\nthis work, we propose a novel multi-behavior recommendation model with\ncascading graph convolution networks (named MB-CGCN). In MB-CGCN, the\nembeddings learned from one behavior are used as the input features for the\nnext behavior's embedding learning after a feature transformation operation. In\nthis way, our model explicitly utilizes the behavior dependencies in embedding\nlearning. Experiments on two benchmark datasets demonstrate the effectiveness\nof our model on exploiting multi-behavior data. It outperforms the best\nbaseline by 33.7% and 35.9% on average over the two datasets in terms of\nRecall@10 and NDCG@10, respectively.\n