Heatmap is a tool to help qualitative researchers identify and curate significant moments from large corpuses of recorded conversation audio.
Each ribbon represents a conversation,and each tile represents a speaker turn. The brightness of each tile represents its 'salience,' which can be scored by several transparent and tunable LLM parameters based on research by Wonjune Kang and Brandon Roy.
Brighter orange and yellow tiles may signal the presence of:
- A coherent narrative
- Emotional significance
- Strong opinions
- Sincere, vulnerable expression
Accordingly, dimmer purple and blue tiles tend to denote the absence of these traits—'filler' such as introductions, affirmations, or single-word interjections.