Research And Implementation Of Visual Analysis Of Subject Evolution Of Scientific Literature

Scientific literature is a written expression of the results and conclusions of scientific researchers after fully observing or studying natural scientific phenomena through experiments.The publication of a large number of scientific documents has enabled the continuous development and improvement of the structure of science.A thorough understanding of the important factors in the development of science can effectively solve environmental,social and technological problems.By analyzing scientific literature data,revealing the subject structure and development history of the subject is essential for understanding the characteristics of the subject,discovering emerging research,and predicting future trends.Focusing on scientific literature,existing research only focuses on the construction of the subject framework or the subject analysis of a single research field,and lacks a complete analysis process for detailed exploration from the top to the bottom of the subject level.In terms of the visual analysis of text topics,the existing methods cannot simultaneously show the correlation between the topics and the temporal changes,and there is less research on crossdomain related topic analysis.Aiming at the above shortcomings,this thesis takes computer science as an example,based on representative conference literature data in the discipline,combined with the actual needs of analysts,and uses visual analysis methods to realize the analysis process of "from discipline to research field to topic".The specific research content and the relevant results are as follows:1)Construct a visual analysis model of the subject evolution of scientific literature.Focusing on the characteristics of literature data,visual analysis tasks are defined from the three levels of overall,individual,and cross-influence.Realize text information mining through technologies such as word vectors and topic modeling.Through a task-driven format,a visual analysis model is constructed,and a variety of visualization methods are proposed accordingly.2)A projection-based document semantic visualization method is proposed to help users understand the macro-structure of the subject.This method uses a combination of scatter plots and contour maps to present the distribution relationship of documents,uses location neighbors to represent the semantic similarity of documents,and uses contours to represent the semantic coverage of the domain.This method supports time-scrolling interaction and helps users understand the evolution of discipline development over time.At the same time,this method adds a partial lens to help users understand the meaning of the view area by adding additional information such as the author and keywords.3)A topic tree visualization method based on flow graph metaphor is proposed to help users understand the topic structure and temporal evolution of the domain.This method integrates the flow visualization method into the tree visualization method of radial layout to realize the theme tree layout.Combined with the clustering-based time slicing algorithm,the theme development stage division is completed.Provides interactive means such as highlighting and brushing to assist users in completing the theme tree exploration.4)Improve the visualization method of Sankey diagram based on correlation,aiming to explore cross-domain subject correlation,such as the application of machine learning technology in the field of visualization.This method is determined by the measurement of topic similarity and the leading-lag relationship between topics,and characterizes the impact between topics.Use the focus + context interactive method to complete the exploration of topic relevance.Based on the above models and methods,a prototype system for visual analysis of the subject evolution of scientific literature was designed and implemented.The system provides a variety of flexible interaction methods based on multi-view linkage to assist analysts in completing tasks such as topic structure analysis,time sequence evolution analysis,and crossimpact analysis.Through case experiment analysis and expert evaluation in computer science scenarios,the effectiveness and usability of the above models and methods are verified.