Research On N-Ary Relation Link Prediction Technology Based On Graph Neural Network

Since the formal proposal of the knowledge graph,it has been widely used in natural language processing fields such as question answering,recommendation system,etc.However,due to the complexity of real world knowledge and the limitation of acquisition,knowledge graphs still have the problem of incompleteness.Therefore,it is of great significance to study the link prediction based on knowledge graph.Currently,researchers mainly focus on link prediction on binary relation,and there is relatively little research on n-ary relation facts.However,in the real world,there is more need to express knowledge through multiple relations.Therefore,link prediction on n-ary relation knowledge graph is the focus and difficulty of research work.Through in-depth and detailed research on the n-ary relation link prediction algorithm,this thesis proposes improvements to the existing problems in current research methods.The main research work is as follows:(1)In order to solve the problem of the relation semantic loss and smoothness of node representation leading to poor relation prediction effect in the prediction process,this thesis uses a global graph to retain the information of nodes in triples and auxiliary information,and at the same time uses the edge aware attention mechanism to distinguish the semantic importance.A node semantic information fusion algorithm is proposed.Through experiments,the effectiveness of this method in dealing with complex semantic relationship prediction is verified.(2)In order to solve the problem of neglecting entity roles leading to poor entity prediction performance,this thesis proposes an n-ary relation link prediction algorithm based on entity role perception.A role perceptron is designed to capture the roles of entities under different relation conditions,and it is injected into corresponding nodes during the prediction process.Through practical verification,it is clear that introducing entity roles in the n-ary relation link prediction process can improve model performance.(3)In order to verify the performance of the proposed algorithm in link prediction,a movie knowledge Question Answering System integrating the above two algorithms was designed and implemented,providing the functions of movie knowledge Q&A and knowledge graph completion.