Research On Temporal Entity Alignment Technology For Knowledge Graph Embedding

The development of internet technology has spawned a large number of platform level application services,such as search,recommendation,query and so on.In order to provide users with more accurate and efficient services,many internet platforms have constructed knowledge graph for different application scenarios.Knowledge graph is a kind of artificial intelligence technology with good organizational structure and strong interpretability,which can organize and express all kinds of information efficiently,so it has been widely used.However,knowledge graph for different network applications is usually constructed independently,so there are problems such as inconsistent expression of the same information and inconsistent update time of data.It is very important for intelligence discovery,case detection,financial risk control and network security to align the time sequence information of the inconsistent entities in the knowledge graph at different times,so as to realize the time sequence information fusion among many application platforms.However,the traditional entity alignment method does not consider the temporal information of entities,and can not effectively fuse the knowledge among the knowledge graph when the knowledge graph changes dynamically and the update state is inconsistent.Therefore,in order to achieve the synergy between dynamic knowledge graph,based on the node centrality algorithm and dynamic knowledge graph embedding algorithm of knowledge graph,this thesis proposes the dynamic knowledge graph embedding algorithm Hy SE based on seed entity and the temporal entity alignment algorithm Dy SEA based on embedding,and implements a person oriented temporal entity alignment system based on Dy SEA.At the algorithm level,the effectiveness of the proposed method is verified by experiments,which can better achieve the temporal alignment fusion between dynamic knowledge graph.The temporal entity alignment system also proves the effectiveness of the proposed method in the actual scene,and solves the problem of temporal entity alignment with characters.The research and innovation of this thesis are as follows:1.Hy SE: dynamic knowledge graph embedding algorithm.In this thesis,we propose a method to optimize the selection of seed entities,which connects different dynamic knowledge graph vector spaces through the selected seed entities.Through experiments,it is proved that the alignment effect of temporal sequence entities can be improved while the negative effects on other tasks such as link prediction can be maintained in a small range;2.Dy SEA: Temporal entity alignment based on Embedding.The embedding method and negative sampling process of dynamic knowledge graph are improved,and two similarity methods are used to align temporal entities.Experimental results show that this method improves the accuracy,recall and F1 value by 27% and 48%,10% and 3.4%,15% and 20%respectively compared with the latest algorithm,which proves the superiority of the proposed method;3.Application system based on temporal entity alignment.This thesis implements an accurate resume system based on temporal entity alignment algorithm.The system contains24756 entities,29 different relationships and 32768 tuple information.It can accurately align the entities of the same person from different data sources and between different people,which proves the effectiveness of the proposed method.