Analysis On Co-word Network Based On Graph Representation Learning

With the rapid development of Internet,the amount of Big Scholarly Data has grown rapidly,which provides a technical foundation and pro-motes the spread of scientific discovery and technological innovation around the world.However,as the amount of data continues to increase,how to extract useful information from scholarly data has become a key problem.As a network constructed from keywords of papers,co-word network provides effective support for scholarly data mining.Graph rep-resentation learning solves the problem that the data cannot be represent-ed by learning graph structure and graph feature,and provides feature representation for further data processing.Co-word network analysis based on graph representation learning can extract useful knowledge from a graph with rich information,construct a knowledge graph for scholarly data,and provide an effective method for exploring technological evolu-tion trends.This thesis focuses on analysis of co-word network based on graph representation learning.The main research work and contribution in-cludes:(1)in order to extract the relation among key words,an algorithm is proposed called Relation Extraction in Co-word network based on Het-erogeneous Graph Convolutional Network and Improved Hierarchical Clustering(HGCN-IHC-REC),which does not rely too much on seman-tics,and captures global structure of co-word network graph.Experi-mental results show that the HGCN-IHC-REC algorithm has better per-formance on accurate relationship extraction rate compared with other benchmark algorithms.(2)Considering the dynamics of the co-word network,an algorithm is presented named Dynamic Graph Representa-tion learning-based Evolution Learning in Co-word network(DGR-ELC).Experiment results indicate the predicted keyword evolution trend and verifies the effectiveness of DGR-ELC algorithm.