Construction And Application Of Academic Knowledge Graph

The 21 st century is an era of rapid development on the Internet,and the data scale within the Internet also shows explosive growth."Cloud Computing","Big Data" and other technical concepts emerge as the times require.However,previous data processing measures only provide search results or some statistical information,lack of visual display of data.Especially in the academic field.In the academic data search websites such as "Baidu Academic" and "Microsoft Academic",you can only directly query cooperative teachers' list of a single teacher.If you want to know the cooperative relationship of teachers in an academic scope,you need to search all teachers in the academic scope in turn.It is not only very cumbersome in operation,but also unable to show this kind of data clearly and completely for users.In view of the above problems,this thesis designed and implemented a Visual Academic Data Search System.It takes teachers as nodes and relationships as edges,and intuitively displays the academic relationships among teachers in different academic scales in the form of relationship graph.The system uses Knowledge Graph technology to store Linked Data,which avoids table join operation of the traditional relational database when querying Linked Data,thus greatly improving the query performance.Therefore,the system can also be called the Academic Knowledge Graph.This thesis mainly carries out the following work to realize the Academic Knowledge Graph.The first step is the acquisition of academic data.Using Protégé tool to design Ontology Layer structure of Knowledge Graph.According to the Ontology Layer definition,using Jsoup crawler framework to obtain the teachers' data of 61 domestic famous universities' official websites.Then using Selenium crawler framework accesses Microsoft Academic official website to obtain the teachers' published paper data.The second step is the processing of academic data.After removing duplicate and invalid data,adding primary key ID,the obtained data of teachers and papers are imported into My SQL database.And the relational data stored in My SQL is converted into RDF triplets by using D2 RQ tool.So as to construct the Data Layer data of Knowledge Map.After importing the Ontology Layer definition and RDF triplets file into Apache Jena framework,the reasoning engine will complete the missing relational data.The third step is the representation and storage of academic data.Because Jena's TDB first query has some serious problems,such as long warm-up time,occupying a lot of memory and so on.The database is constructed by using RDF standard triples.So the ontology definition is complex and query efficiency is less than Graph Database with simple structure.Therefore,Neo4 j is used to replace Jena's TDB for knowledge representation and storage.The fourth step is the visualization of academic data.The back-end function of the Academic Knowledge Graph is realized by using springmvc framework and Neo4 j.The academic relationship data obtained from the back-end is input into Gephi software to calculate the position coordinates displayed on the visual page for the teacher entities.Then the academic relationship data is packaged into Json format and sent to front-end's Apache Echarts framework,which supports the presentation of academic relationships for users by the graph of node and edge.After the completion of the system development work,the experiment compares the vectorization effect of four popular Knowledge Graph Embedding Algorithms on the storage of triplet in the Academic Knowledge Graph.Link prediction is used as an evaluation index to analyze the embedding algorithm which is more suitable for the vectorization of the system data set.