Research And Application Of Knowledge Fused Hierarchical Topic Model

With the continuous development of the informatization of scientific research management in domestic university,the amount of all kinds of scientific research documents increased rapidly.How to effectively organize and utilize these scientific research documents is an urgent problem.The hierarchical topic model can mine the implied topics and hierarchical relations between topics from document collections,which can deeply understand and analyze the data.However,these unsupervised models,which do not incorporate any prior knowledge,tend to produce weak topic hierarchy.To solve this problem,this thesis deeply researched the knowledge fused hierarchical topic model,and explored its application on the scientific research management platform.For the document collections with title information,this thesis proposed a new hierarchical topic model,called Knowledge-fused Hierarchical Latent Dirichlet Allocation(KHLDA),which can integrate the knowledge of word correlation contained in the documents and effectively utilize the guiding effect of the document title on the document topics in the process of hierarchical topic modeling of document collection.We evaluated the effectiveness of hierarchical topic extraction on the scientific research document collection using the proposed model.The experimental result shows that KHLDA makes full use of the knowledge in the documents,which can effectively improve the ability of hierarchical topic extraction.Moreover,this thesis applied the proposed model to the scientific research management platform.Based on the hierarchical topic structure extracted out from the scientific research documents by the model,we implemented the organization,retrieval and analysis of scientific research documents from the perspective of hierarchical topics,which includes the functions of visualization of hierarchical topics,document search and recommendation based on hierarchical topics and evolution of hierarchical topics.The proposed model effectively mines the latent information in research documents and enriches the methods of understanding and analyzing research documents,has high value in research and application.