Research On University Academic Knowledge Graph Construction And Its Application Based On Multi-source Data

Knowledge graph is a large-scale semantic network proposed by Google,which aims to describe various entities,concepts and relations existing in the real world.As an important knowledge representation method in the era of big data,knowledge graph has become one of key technologies of artificial intelligence.It is widely used in social networks,intelligent recommendation,in-depth question answering and other fields.Academic data is an important type of data in universities,which can reflect the comprehensive scientific research ability and innovation ability of universities.However,academic data has the characteristics of large volume,multiple sources,diversity and dynamics.How to organize and manage academic data to analyze the data has become an important research content.To this end,this dissertation studies the construction of university academic knowledge graph and representation learning models.The main research contents and results include:(1)Aiming at the characteristics of multi-source and diversity of academic data formats,a university academic knowledge graph based on multi-source data is constructed.Extract knowledge elements such as entities,relations,and attributes from various types of data sources such as unstructured data,semi-structured data and structured data,and organize them into triples,and then fuse the triples into a unified knowledge graph to provide a structured knowledge base for data analysis through knowledge fusion algorithm.(2)Aiming at the problems of data sparsity and complexity of knowledge graph,a representation learning model of knowledge graph with semantic vectors(Trans V)is proposed.Introduce a text corpus and knowledge graph context to construct semantic vectors for entities and relations,and design a semantic matrix for each relation,which deeply expands knowledge graph from a semantic perspective.The new training function is designed to help improve the accuracy of knowledge graph representation learning.Experiments show that,compared with the existing translation models,Mean Rank is reduced by 66 and 162 on average,and Hit@10 is increased by 20% and 19%on average on FB15 K and WN18 datasets.(3)A prototype system for university academic data analysis based on knowledge graph is developed,including four modules: teacher portrait,scholar cooperation,number of papers published,and research directions of scholars.The teacher portrait module describes scholars from five aspects: scholars' basic information,published papers,participated vertical projects,horizontal projects,and patents.Finally,the scholars are analyzed and displayed in a visual way.