Multi-modal Graph Attention Network for Video Recommendation

Abstract

In view of the problems of cold start and data interaction in recommendation systems, and most current recommendation algorithms ignore the diversity of data types, the combination of multimodal data and knowledge graph is bound to improve the pertinence of video recommendation. In this paper, we propose Multi-modal Knowledge Graph Attention Network (MMKGV) model, and all the entity nodes of the knowledge graph are innovatively introduced into multimodal information. The high-order recursive node information dissemination and information aggregation are carried out on the multimodal knowledge graph through the graph attention network. In the model, the triplet function of the knowledge graph is used to construct the triplet inference relationship, and the vector representation generated by the final aggregation is used for recommendation. Through extensive experiments on two public datasets TikTok and Kwai, the results show that the MMKGV can effectively improve the effect of video recommendation compared with other comparison algorithms.