PyramidTags: Context-, Time- and Word Order-Aware Tag Maps to Explore Large Document Collections

Area of Science:

• Information Visualization
• Natural Language Processing
• Data Analysis

Background:

• Exploring large text collections is challenging due to limited initial information.
• Existing visualization methods offer limited insights into temporal and semantic document relationships.
• Stakeholders across various domains require effective tools for corpus analysis.

Purpose of the Study:

• Introduce PyramidTags, a novel visual approach for summarizing large text collections.
• Enhance the representation of text collections by incorporating temporal evolution and semantic relationships of tags.
• Provide analysts with an interactive visual starting point for corpus exploration.

Main Methods:

• Developed PyramidTags, a novel visualization technique for text summarization.
• Incorporated temporal evolution and semantic relationships of tags into the visualization.
• Applied PyramidTags to a large corpus of hundreds of thousands of web-crawled news reports.

Main Results:

• PyramidTags generates time- and context-aware layouts for tag visualization.
• The approach preserves the inherent word order of important tag pairs.
• Enabled interactive exploration of tag relationships and document retrieval.

Conclusions:

• PyramidTags provides an effective visual starting point for exploring large text collections.
• The method improves understanding of temporal dynamics and semantic connections within corpora.
• Facilitates interactive discovery of key terms, phrases, and underlying narratives in text data.