Research On The Construction And Application Of Temporal Academic Knowledge Graph

Academic knowledge graphs are frequently applied to knowledge service scenarios like literature retrieval,knowledge question and answer sessions,and discipline analysis.Currently,academic knowledge graphs are predominantly static graphs that cannot accurately obtain time-series knowledge,such as the current research fields of authors and the directions of currently included papers in journals,as well as other time-series information,and have limitations in scenarios such as knowledge quizzes and time-series recommendations.This thesis introduces the temporal knowledge graph technology,extracts temporal data from academic data,constructs the Temporal Academic Knowledge Graph(TAKG),proposes a recommendation model based on the temporal academic knowledge graph,and applies it to tutor recommendation and journal recommendation to address the aforementioned issues.In conclusion,we extract the scholar genealogy graph from the Temporal Academic Knowledge Graph and propose a reviewer review model based on the scholar genealogy graph to handle the relationship conflict issue in the paper review process.The major work of this thesis is as follows:(1)The development of a temporal academic knowledge graph.First,academic entities and temporal relationships of the temporal academic knowledge graph are defined,including academic entities such as papers,authors,and institutions,and temporal relationships such as writing time,publication time,and working time;second,temporal data are extracted according to the definition using academic metadata and stored in the form of quaternion;and finally,the quaternion data is used to generate the temporal academic knowledge graph,which is then saved in the Neo4 j graph database.(2)The Temporal Academic Knowledge Graphs for Recommendation(TAKGRec)recommendation model is proposed.First,we build a temporal academic knowledge graph embedding model to map academic entities and temporal relationships into embedding vectors.Next,we stitch and fuse the embedding vectors and their weighted vectors by DNN to obtain feature fusion vectors.Experiments demonstrate that the TAKGRec model outperforms the comparison model across all evaluation indices,confirming that the temporal academic knowledge graph is more effective for the recommendation task.(3)A Reviewer Review Based on the Scholar Genealogy(R2SG)is proposed.First,we extract the scholar genealogy graph by leveraging the temporal academic knowledge graph,then we obtain the entity embedding representation by fusing the temporal information with the graph embedding model,and finally,we filter the first-order neighbouring nodes and the second-order feature similar nodes to obtain the set of reviewers with conflicting relationships.The experimental results demonstrate that the R2 SG model efficiently excludes reviewers with contradictory ties,demonstrating the methodology’s efficacy.