Conference Evaluation And Knowledge Representation Based On Big Scholarly Data

Academic conferences are an important channel for knowledge dissemination.With the increase in the number of academic conferences and the complexity and variety of scholarly data,researchers pay more attention to the quality of conferences.There is an increasing demand in academia for an objective and professional evaluation system for academic conferences.How to help researchers quickly and efficiently understand academic conferences,accurately represent academic knowledge,and broaden the application scenarios of academic data has become a current research hotspot.Firstly,this thesis proposes an attractiveness-based method for evaluating the impact of academic conferences.The attractiveness of conferences is quantified by two indicators,freshness and tightness.Meanwhile,the H5-index of conferences,the H-index of authors and the self-citation rate of conferences are also selected to comprehensively evaluate the influence of the conference from five dimensions.Based on the experimental results,the changes of rankings and scores of 10 conferences in the same field within 5 years were analyzed.The algorithm provides a reference for researchers to submit manuscripts.Secondly,this thesis constructs an academic conference graph based on the public big scholarly data and stores it in the Neo4 j database.It can effectively organizing the knowledge of academic conferences and provide data support for academic conference evaluation.Based on the conference graph,functions such as conference retrieval,conference recommendation,and conference analysis are also implemented.Finally,in order to more accurately represent academic entities and relationships,this thesis proposes a network motif-based multi-relational knowledge graph representation learning algorithm,called MED-GCN.It is mainly based on graph convolutional networks,and the core idea is to learn the information contained in neighbor entities and relations,and aggregate this information to update the central entity representation.In this model,the high-order structural information of the graph is considered,the importance of the relationship is quantified by the motif edge degree,different learning weights are assigned to different neighbor entities,so as to optimize the representation of entities and relationships.Through experiments on existing classic datasets and real datasets constructed by ourselves,it is show that the proposed algorithm is superior to other existing knowledge representation algorithms in link prediction and entity classification tasks.