CTAN: Collaborative Tag-Aware Attentive Network for Recommendation

Abstract

Graph-based methods are one of the effective means to solve the data sparsity and cold start problems. They can not only ensure the accuracy of the recommendation results but also have a certain degree of interpretability. However, it is worth mentioning that existing Tag-aware Recommendation Systems (TRS) rely on tag-aware features for recommendations, which is insufficient to alleviate the problems of sparsity, ambiguity, and redundancy brought about by tags, thereby hindering the recommendation performance. Therefore, an end-to-end framework is proposed in this paper. Firstly, a Collaborative Tag Graph (CTG) is designed to handle the problem of extracting heterogeneous semantic information. Secondly, an attention mechanism-based Graph Neural Network (GNN) is introduced to distinguish the importance of neighboring nodes, utilizing the information propagation mechanism of GNN to update node representations. Moreover, a dual interaction aggregator is used to aggregate the neighboring nodes and self-node characteristics. Experiments on three benchmark datasets demonstrate that CTAN outperforms the state-of-the art methods consistently.

Supported by Industry and Information Technology Ministry Industrial Promotion Center Project, Intelligent Analysis Engine for Industrialization of Innovation Achievements in the Manufacturing Industry.