Research And Application Of Recommendation Technology Based On Hypergraph Completion

With the powerful modeling ability of knowledge graph to the real world,the recommendation technology based on knowledge graph shows superior performance,but due to the complexity of the relationship in the real world,the knowledge graph is difficult to represent the common multivariate relationship in the real world,resulting in the lack of semantic connotation,which in turn leads to the limitation of recommendation ability.In addition,the existing knowledge graph has the absence of relationships or entities,resulting in incomplete knowledge.To solve the above problems,the knowledge graph introduces the concept of hypergraph to model the high-order complex structures in multivariate relationships,and realizes the knowledge completion,thereby greatly enriching the semantics of the knowledge graph and further improving the recommendation ability.This thesis proposes a local information aggregation model based on semantic perception and a hyperedge semantic association model based on graph learning for knowledge hypergraphs,then verifies the application of the model in recommendation technology.The main work of this article is as follows:(1)Aiming at the problems of existing hypergraph models such as the loss of internal structure information of multivariate relations and incomplete semantic perception,a local information aggregation method based on semantic perception is proposed.This thesis,with the facts of multivariate relations represented by triples and auxiliary keyvalue pairs,perceives the semantic expression of relationships under different head-tail entity combinations in triples and captures the local features of entities based on their specific structure in auxiliary key-value pairs;finally,uses the graph neural network to aggregate the multivariate neighborhood information of relational facts.Experiments show that the model has better modeling ability in multivariate relational facts,thus verifying the effectiveness of the model.(2)Aiming at the problem that existing models rely too much on entity and relationship feature vector representations when transferring information between facts,this thesis proposes a method based on hyperedge semantic association approach based on graph learning.In this thesis,the knowledge hypergraph is transformed into the corresponding directed line graph,thereby jumping out of the fact of multiple relationship itself,using the graph neural network to learn the macro-structure information of the knowledge hypergraph,and capturing the semantic association between hyperedges,and more effectively mining the potential semantic links between multi-relational facts.Experiments show that the model has better inference on real-world datasets.(3)By constructing a recommendation dataset in the form of a knowledge hypergraph,this thesis compares experiments with the existing recommendation models with the feasibility of the hypergraph completion model verified in the recommendation technology and the accuracy of the recommendation results further improved.Finally,a movie recommendation system based on the hypergraph completion model is designed and implemented.