Conference material: "Proceedings of the International Conference on Computer Graphics and Vision “Graphicon”"
Authors:Charnine M., Sokolov E., Klokov A.
Semantic approach to visualization of evolution dynamics of topic trends in space of scientific publications using t-SNE and web-based 3D graphics
This paper describes a semantic approach to visualization of 3D cyberspace of Artificial Intelligence (AI) publications and their topic trends evolution using web-based 3D graphics. The purpose of research is to group AI publications with same subject into clusters for further visualization of topic trends dynamics. An unsupervised method and algorithm for visualizing the dynamics of topic trends by generating a time series of 2D and 3D semantic visual maps with predictive information is described. The method includes semantic similarity measure and citation prediction for documents, topic modeling and clustering, dimensionality reduction, virtual reality technology, representation of dynamics using time filters. As an example of particular implementation, the method is demonstrated on AI collection data using technologies of neural network prediction, LDA clustering, t-SNE dimensionality reduction, WebVR visualization. Cluster dynamics associated with scientific trends is analyzed. The growth in number of clusters and their consolidation during the period from 1954 to 1993 is demonstrated. It is shown that 3D visual map better preserves articles similarity and high-dimensional clusters structure than 2D visual map. The proposed cyberspace implemented by WebVR and interactive 3D graphics can be considered as a dynamic learning environment that is convenient for discovering new significant articles, ideas and trends.
virtual reality, web-based 3D graphics, WebVR, scientific papers, topic modeling, dynamics of topic trends, semantic similarity, visualization, visual map