Effective analysis of large text collections remains a challenging problem given the growing volume of available text data. Topic model, which is a type of statistical models for discovering the hidden thematic information in a collection of documents, has been widely used to analyze massive text data. In this thesis, we develop a visual analytic system, VISTopic, which integrates a state-of-the-art topic model, hierarchical latent tree model (HLTM), with interactive visualizations to help users make sense of large document collections. To exploit the hierarchical property of this model, VISTopic utilizes a Sunburst diagram to present the topic organization, customizes the layout of tag clouds to present the topic semantics and merges the two visualizations into a single topic vie...[ Read more ]

Effective analysis of large text collections remains a challenging problem given the growing volume of available text data. Topic model, which is a type of statistical models for discovering the hidden thematic information in a collection of documents, has been widely used to analyze massive text data. In this thesis, we develop a visual analytic system, VISTopic, which integrates a state-of-the-art topic model, hierarchical latent tree model (HLTM), with interactive visualizations to help users make sense of large document collections. To exploit the hierarchical property of this model, VISTopic utilizes a Sunburst diagram to present the topic organization, customizes the layout of tag clouds to present the topic semantics and merges the two visualizations into a single topic view. In addition, VISTopic also contains an evolution view to reveal the trend of topic with a ThemeRiver and a document view to provide the detailed information for certain topical documents with a Bubble chart. To demonstrate the effectiveness of VISTopic, we have applied VISTopic to exploring and analyzing the corpus of IEEE VIS conference and obtained some interesting findings.