Contrastive Meta Learning with Behavior Multiplicity for Recommendation

A well-informed recommendation framework could not only help users identify interested items, but also benefit the revenue of various online (e. g. , e-commerce, social media). Traditional recommendation models assume that only a single type of interaction exists between user and, and fail to model the multiplex user-item relationships from multi-typed behavior data, such as page view, add-to-favourite and purchase. While recent studies propose to capture the dependencies across different types behaviors, two important challenges have been less explored: i) Dealing with sparse supervision signal under target behaviors (e. g. , purchase). ii) the personalized multi-behavior patterns with customized dependency. To tackle the above challenges, we devise a new model CML, Meta Learning (CML), to maintain dedicated cross-type behavior for different users. In particular, we propose a multi-behavior learning framework to distill transferable knowledge across types of behaviors via the constructed contrastive loss. In addition, capture the diverse multi-behavior patterns, we design a contrastive meta to encode the customized behavior heterogeneity for different users. experiments on three real-world datasets indicate that our method outperforms various state-of-the-art recommendation methods. Our studies further suggest that the contrastive meta learning paradigm great potential for capturing the behavior multiplicity in. We release our model implementation at: : //github. com/weiwei1206/CML. git.