| With the rapid development of the mobile Internet,the amount of data in the network is increasing exponentially,which leads to the problem of information overload and makes it difficult for users to find what they are really interested in from the massive products or services on the Internet.In order to improve user experience and enterprise economic benefits,the recommendation system came into being.The core of an effective recommendation system is to accurately model user preferences based on user history interaction and to recommend items to users according to user preferences.Although modern recommendation systems are thriving,they are always affected by data sparsity and cold start problems.As a kind of graph neural network,hypergraph neural network can make better use of high-order association information in data,thus effectively alleviating the problems of data sparseness and cold start.This article designs corresponding recommendation models based on hypergraph neural network for three different recommendation task scenarios respectively in order to achieve better recommendation effect.The main research contents of this article are as follows:(1)In order to better model the higher-order association between entities on the knowledge graph and combine it with the recommendation task,we propose a knowledge-aware hypergraph neural network model.The model first constructs hyperedges based on user history interaction and knowledge graph,then constructs knowledge-aware hypergraph for user and item respectively according to the generated hyperedges,and then uses knowledgeaware hypergraph convolution method to learn the final embedded representation of corresponding user or item on the hypergraph.(2)In order to better capture the global high-order association information of different items in and between sessions and integrate it with the local association information between items in session,we propose a session-based recommendation model based on hypergraph attention network.Firstly,the gated graph neural network is applied to capture the local association information on the local subgraph of the session,then the hypergraph attention network is applied to capture the global association information on the global hypergraph composed of the session hyperedge,and then the two representations are fused and the session sequence information is captured by the attention mechanism.(3)In order to better model the high-order association between users and items and combine it with collaborative filtering information,we propose a collaborative filtering recommendation model based on hypergraph attention network.The model uses hypergraph attention convolution method and graph convolution method to capture high-order association information and collaborative filtering information between nodes on the hypergraph and bipartite graph constructed from user history interaction respectively,and the two kinds of information are fused and used for recommendation. |
