Research On Knowledge Fusion For Knowledge Graph Construction

With the rapid popularization of the Internet,how to effectively organize,utilize and mine the knowledge hidden behind the data has become a new challenge.Knowledge Graph,which describes the entity and semantic relations existing in the objective world and presents them intuitively in the form of graph structure,provides users with structured knowledge and gradually attracts the general attention of academia and industry.Therefore,how to construct a knowledge graph has become a hot topic for scholars.At the same time,multiple sources and heterogeneous knowledge often have problems such as duplication,diverse semantic ambiguities,and uneven quality.In order to build a highquality knowledge graph,knowledge fusion is an indispensable key link.Entity alignment and entity linking are two important subtasks of the knowledge fusion task.The existing methods still have obvious shortcomings.The main manifestation is that the existing entity alignment methods cannot fully utilize the attribute information contained in the attribute triples,and the existing entity link methods ignore the structural information of the knowledge graph and leads to low link accuracy,so it is worthy of further research and improvement.This paper applies the knowledge graph to the field of movie and television,deeply studies the key technologies of knowledge fusion involved in the construction of multilevel film and television knowledge graphs,the two important sub-tasks of knowledge fusion are entity alignment and entity linkage.The work contents of this paper are as follows:(1)Aiming at the shortcomings of existing entity alignment methods,this paper proposes a Nov EA model.Firstly,this model makes full use of the attribute triples and relationship triples in the knowledge graph.Then this model prioritizes the attributes according to the domain characteristics of the knowledge graph.Finally,from the perspectives of structure and attributes,a binary regression model is used to analyze entities.The similarity between them is measured,and the weights between relationships and attributes are dynamically adjusted,which further enhances the effect of entity alignment.Experiments show that compared with other models of the same task,the alignment accuracy of the Nov EA model proposed in this paper is significantly improved.(2)Aiming at the shortcomings of existing entity linking methods,this paper proposes a knowledge graph entity linking model based on multi-dimensional granularity.Based on the traditional entity linking method,this model focuses on the neighborhood information of candidate entities in the knowledge graph.In the sorting module of this method,the similarity calculation between the entities to be linked and the candidate entities are carried out from multiple dimensions,such as entity name,entity description and the neighborhood of the graph entity node,so as to select the best matching candidate entity.Experiments prove that the accuracy of the multi-dimensional granular entity link model proposed in this paper is better than other models of the same task.(3)Apply the knowledge fusion method of this article to the field of film and television,and construct a multi-level film and television knowledge graph.Firstly,the corresponding single-level knowledge graph is constructed based on the three different entity levels of movie entity,figure entity and organization entity.Secondly,due to the relevance between different entity levels,this paper integrates single-level knowledge graphs constructed by different entity levels to construct a high-quality,multi-level film and television knowledge graph with comprehensive content and three-dimensional structure.Finally,the collection of fragmented film and television knowledge is realized,and at the same time,systematic knowledge is formed.