Deep Global and Local Matching Network for Implicit Recommendation

Wei Yang, Yiqun Chen, Jun Sun, Yan Jin

Abstract: In real recommendation scenarios, users often have implicit behaviors including clicks, rather than explicit behaviors. In order to solve the matching problem under implicit data, many researchers have proposed methods based on neural networks, mainly including representation learning and matching function learning methods. However, these methods do not take into account the diverse preferences of users, and there is no fine-grained modeling matching relationship. In this paper, we consider the matching problem from a new perspective and propose a novel deep global and local matching network (DeepGLM) model. In detail, DeepGLM introduces multi-aspect representations to express the user’s various preferences, and calculates the global matching degree between user and item through the hierarchical interactive matching module. Then, the attention mechanism is adopted to calculate the local matching relationship based on feature interactions. In addition, the gating mechanism is used to control the effective transmission of global and local matching information. Extensive experiments on four real-world datasets show significant improvements of our proposed model over the state-of-the-art methods.