TextFlow: towards better understanding of evolving topics in text

Area of Science:

• Natural Language Processing
• Data Visualization
• Information Retrieval

Background:

• Analyzing topic evolution in text data is challenging, with prior work focusing mainly on individual topics.
• Understanding multi-topic evolution is crucial for comprehensive text data analysis.

Purpose of the Study:

• To introduce TextFlow, a system integrating visualization and topic mining for analyzing multiple topic evolution patterns.
• To extend existing techniques for extracting multi-level features of topic evolution.

Main Methods:

• Extending analysis techniques to extract topic evolution trends, critical events, and keyword correlations.
• Designing a coherent visualization with three new components to represent complex relationships.
• Implementing an interactive system where topic mining and visualization communicate to refine analysis.

Main Results:

• TextFlow effectively extracts three-level features: topic evolution trend, critical event, and keyword correlation.
• A novel visualization effectively conveys complex relationships between these features.
• The interactive system allows progressive refinement of analysis and data insights.

Conclusions:

• TextFlow demonstrates effectiveness and usefulness in understanding major topic evolution patterns in time-varying text data.
• The integration of visualization and topic mining offers a powerful approach for complex text data analysis.