Visualization of Subtopics of the Thematic Document Collection Using the Context-Semantic Graph

Abstract

An approach for visualization of nested topics within large collections of documents is proposed. The approach is based on set of parameters: information entropy, Kullback-Leibler divergence, Ginzburg algorithm, similarity the distributions of keywords and key phrases in the documents with Bernoulli's theoretical distributions. The results of comparisons of our approach with implementations based on TF-IDF approaches are presented.