Academic Knowledge Graph Based Topic Evolution Analysis Method

Academic papers are an important source of information for scientific researchers in various industries to find new directions or ideas of scientific research,to understand the research achievements and development process in various fields,and to know the research hotspots and research progress.When researchers carry out literature information retrieval,the annotation of related topics,the display of topic hot and topic evolutionary trend between topics can provide great help for researchers to obtain the above information.However,current academic literature retrieval system and topic evolution analysis method have the following problems.Firstly,the ranking of searching results is often limited in text matching,and it fails to optimize the ranking using topics users are interested in.Secondly,the construction of the academic knowledge graph is still imperfect,lacking the representation of topics and the quantification of the degree of relationship among entities.Thirdly,the general research on topic evolution analysis mainly relies on the similarity of documents on the text level as the basis of judging whether the documents are related or not.In the scenario of academic papers,the relationship between entities such as citation and writing embodied in academic knowledge graph is not utilized.In order to solve these problems,this paper proposes a topic evolution analysis method based on academic knowledge graph.Using structured paper data to construct academic knowledge graph,training improved topic model to augment the knowledge graph by adding entities and assigning weight to relationship,and giving different similarity calculation methods for different types of entities.Dividing the time knowledge graph on time dimension,modeling and analyzing the topic hot and evolutionary trend in various research fields based on time-sliced knowledge graph.The main research contents of this subject are as follows:(1)Propose a framework of topic evolution analysis method based on academic knowledge graph.This paper proposes a framework of topic evolution analysis method based on academic knowledge graph.Using structured paper data as input,after topic model training,knowledge graph construction and augmentation,entity similarity measuring and topic hot and inter-topic evolution trend analyzing,the analysis results are presented to users by academic search engines or retrieval tools.(2)Construct an improved topic model and augment the academic knowledge graph.This paper constructs an improved topic model,allowing topics to be generated by other academic entities besides papers.Using the improved topic model,topics are added to the academic knowledge graph as additional entities,and the weight of the relationship in the knowledge graph is quantified.Different entity similarity measuring methods are given for different types of entities.(3)Analyze topic trend based on time-sliced knowledge graph.This paper constructs a time-sliced knowledge graph.Based on the topic entities and weighted relationship in the time-sliced knowledge graph,we analyze the temporal change of topic hot and influence,and the evolutionary relationship between different topics.(4)Develop a prototype system of academic paper retrieval tool.By combining the similarity between query text and papers in topic distribution,we re-rank the retrieval results obtained by text matching.The relevant academic entities are recommended according to the results of entity similarity measuring.According to the above method framework,a prototype system of academic paper retrieval tool is developed,and we introduce the application scenario with the use cases of paper retrieval and topic evolution trend viewing.Finally,we highlight the innovative features of this paper and verify the validity of the proposed method through experimental comparison and analysis.