Recommendation Algorithm for Multi-Task Learning with Directed Graph Convolutional Networks

Abstract

As an important branch of machine learning, recommendation algorithms have attracted the attention of many experts and scholars. The current recommendation algorithms all more or less have problems such as cold start and single recommended items. In order to overcome these problems and improve the accuracy of personalized recommendation algorithms, this paper proposes a recommendation for multi-task learning based on directed graph convolutional network (referred to as MTL-DGCNR) and applies it to recommended areas for e-commerce. First, the user’s micro-behavior is constructed and converted into directed graph structure data for model embedding. It can fully consider the embedding of first-order proximity nodes and second-order proximity nodes, which can effectively enhance the transformation ability of features. Secondly, this model adopts the multi-task learning method, and uses knowledge graph embedding to effectively deal with the one-to-many or many-to-many relationship between users and commodities. Finally, it is verified by experiments that MTL-DGCNR has a higher interpretability and accuracy in the field of e-commerce recommendation than other recommendation models. The ranking evaluation experiments, various training methods comparison experiments, and controlling parameter experiments are designed from multiple perspectives to verify the rationality of MTL-DGCNR.