Conch: Competitive Debate Analysis via Visualizing Clash Points and Hierarchical Strategies

Area of Science:

• Computational Social Science
• Human-Computer Interaction
• Argumentation Theory

Background:

• Manual analysis of debate transcripts is inefficient for identifying argumentative strategies and context.
• Unstructured debate data hinders the reconstruction of semantic meaning and logical flow.

Purpose of the Study:

• To introduce Conch, an interactive visualization system for systematic competitive debate analysis.
• To enhance understanding of both the content and dynamics of debates.
• To improve debaters' performance through effective analysis tools.

Main Methods:

• Developed a novel parallel spiral visualization to track clash points and participant interactions.
• Utilized large language models with specific prompts to automatically detect key debate elements (clash points, disagreements, viewpoints, strategies).
• Conducted case studies with real-world debates and user studies to evaluate the system.

Main Results:

• Conch effectively analyzes multidimensional aspects of debates, including clash evolution and interaction patterns.
• Automated identification of debate elements by LLMs provides comprehensive contextual understanding.
• User studies confirmed the effectiveness and usability of Conch for competitive debate analysis.

Conclusions:

• Conch offers a systematic and efficient approach to analyzing competitive debates, overcoming limitations of manual methods.
• The integration of visualization and LLMs provides valuable insights for debaters.
• The system demonstrates potential for improving argumentative skills and debating performance.