Commit dd6d137
committed
Add new readability and conversational metrics to evaluation module
- Added three new readability metrics:
* Coleman-Liau Index: Grade level estimation using character counts
* Linsear Write: Grade level based on easy vs difficult words
* Dale-Chall Readability Formula: Difficulty assessment using familiar word list approximation
- Added five new conversational features:
* Turn-taking ratio: Conversation balance using normalized Shannon entropy (0=monopolized, 1=balanced)
* Question rate: Percentage of turns containing questions
* Lexical diversity: Type-token ratio measuring vocabulary richness
* Back-channel rate: Percentage of minimal response turns (acknowledgments)
* Filler word density: Percentage of filler words like "like", "you know", "basically"
- Refactored metrics into semantic classes:
* ConversationalFeatures: Dialogue structure and interaction dynamics
* ReadabilityScore: Text complexity and reading difficulty
- All metrics support speaker filtering and flexible feature selection1 parent 011def5 commit dd6d137
2 files changed
+579
-54
lines changed
0 commit comments