Cascading Residual Graph Convolutional Network for Multi-Behavior Recommendation

Multi-behavior recommendation exploits multiple types of user-item interactions, such as view and cart , to learn user preferences has demonstrated be an effective solution alleviate the data sparsity problem faced by traditional models that often utilize only one type interaction for recommendation. In real scenarios, users take a sequence actions interact with item, in order get more information about item thus accurately evaluate whether fits their personal preferences. Those behaviors obey certain order, importantly, different reveal or aspects towards target item. Most existing multi-behavior methods strategy first extract from separately then fuse them final prediction. However, they have not exploited connections between Besides, introduce complex model structures parameters behaviors, largely increasing space time complexity. this work, we propose lightweight named C ascading R esidual G raph onvolutional N etwork (CRGCN short) recommendation, which can explicitly exploit into embedding learning process without introducing any additional (with comparison single-behavior based model). particular, design cascading residual graph convolutional network (GCN) structure, enables our continuously refining embeddings across behaviors. The multi-task method is adopted jointly optimize on Extensive experimental results three real-world benchmark datasets show CRGCN substantially outperform state-of-the-art methods, achieving 24.76%, 27.28%, 25.10% relative gains average terms HR@K (K={10, 20, 50, 80}) over best baseline datasets. Further studies also analyze effects leveraging multi-behaviors numbers orders performance.