Research On Knowledge Structure Discovery And Its Evolutionary Mechanism Based On Topic Model

The intersection and generalization of many scientific researches have resulted in a complex situation of interlaced researches in various fields.Diversified research contents bring us some troubles in understanding and mastering the internal structure of knowledge and its evolutionary mechanism,and the contradiction between vast knowledge and limited personal energy is inevitable.Especially for scholars who are new to a certain field,it often takes a lot of work to understand the knowledge structure and evolution mechanism of the field comprehensively and quickly.Scattered knowledge points and unstructured information are not conducive to the formation of knowledge structure,which also hinders the further development of evolutionary research.Aiming at this problem,this paper proposes a hierarchical method for the discovery of the structure of scientific knowledge,and on this basis carries out the evolution and prediction of knowledge topics,and further analyzes the research status and development trend of the scientific field.The main work of the paper is divided into the following two aspects: Discovering the hierarchical structure of scientific knowledge.On the basis of summarizing and analyzing the current mainstream knowledge structure discovery methods,the LDA topic model is used to design the hierarchical knowledge structure discovery framework.The framework is divided into data layer,logic layer and presentation layer from top to bottom.Firstly,in the part of literature preprocessing,"double conjunctions" in corpus are extracted as the input of the theme model modeling file,which greatly improves the representability of the topic and makes the obtained knowledge topic more consistent with general cognition and easy to understand.Secondly,by introducing the average similarity between subjects,the termination layer of knowledge structure is determined on the basis of ensuring the distinctiveness between subjects.In addition,an algorithm for automatic screening threshold values in the "document-topic" probability matrix is designed to balance the relationship between topic quality and document scope and help knowledge topics determine the scope of the subsets of the underlying literature.Finally,a hierarchical tree of knowledge structure is generated for the visualization of the results.Chinese literatures in the field of "cloud computing" in recent ten years were selected as data sets for experimental verification.By comparing with traditional hierarchical methods,the methods proposed in this paper were greatly improved in terms of document membership,time complexity,single-layer topic differentiation and inter-layer topic inheritance.Research on the evolution mechanism of knowledge topic.It mainly includes the stable evolution and prediction of the topic based on markov and the heat evolution and prediction of the topic based on hidden markov.The former is based on the analysis of the evolution law of topics,and takes the similarity matrix between topics as the transition probability matrix in markov model to analyze the evolution and transition process of topics and predict the steady-state distribution of topics.The latter uses the three indexes of the downloads,citations and posts under the theme to comprehensively judge the heat of the topic.The concept of topic purity was proposed,heat transfer matrix and confusion matrix were constructed,and the models were modeled on the basis of hidden markov theory.Finally,we take the knowledge topic in the hierarchical knowledge structure of the "cloud computing" domain as the object for model validation.Compared with the gray model,our model has a smaller average relative error and higher prediction accuracy.The hierarchical knowledge structure discovery method constructed in this paper can discover the hierarchical knowledge structure and research topic in the field,and solve the related problems of its evolution prediction.It can reflect the research status and evolution mechanism of the field more accurately,and provide reference for the development decision of the field.