Recommender System Based On Multi-modal Knowledge Graphs

With the rapid development of information technology and the internet,the amount of information that people can access is increasing exponentially.However,each user's time is limited.As the amount of information grows,the efficiency of users' consump-tion of information slows down.This phenomenon is called Information Overload,and it will cause users to be unable to find the information they want in a short period of time.The recommendation system was born to alleviate the problem of information overload.However,users have consumed very little information in various Internet platforms.so the recommendation system will face problems such as data sparsity.In order to solve these problems,researchers usually use some external information to assist the recommenda-tion algorithm.Among them,the knowledge graph based recommendation algorithm has attracted the attention of researchers.However,most of these works did not make use of multi-modal information such as texts and images that are common in life,which will affect users' choices.Multi-modal knowledge graph(MKGs)is a technology that introduces multi-modal information into the knowledge graph,so it can be used as the knowledge source of the rec-ommendation system,which can naturally make the recommendation system based on the knowledge graph use the multi-modal information.However,as multi-modal knowledge graph has only been introduced in recent years,there are only limited research works in this direction.Therefore,exploring the combination of the multi-modal knowledge graph and recommendation system will be a promising work.To explore this new field,this pa-per proposes a Multi-modal Knowledge Graph Attention Network(MKGAT),which can use the information of the multi-modal knowledge graph to improve the effectiveness of the recommendation algorithm.To the best of our knowledge,this is the first work that incorporates multi-modal knowledge graph into recommender systems.In addition,the MKGAT proposed in this paper innovatively transforms multi-modal information fusion into graph information prop-agation,which further enrich the representation of items and users in the recommendation system.Finally,this paper constructs two recommendation system datasets containing multi-modal knowledge graphs,and conducts detailed experiments on these two datasets.The results show the effectiveness of the MKGAT.And this paper analyzes the modules in MKGAT in the ablation experiment,which can show that MKGAT can efficiently use the multi-modal knowledge graph to improve the performance of the recommendation algorithm.