First run downloads two small models (~90 MB), then they are cached. Untick for instant keyword mode.
Mentions by theme (number of reviews)
negativeneutralpositive
Pain points ranked by priority
Priority score = % of reviews with a negative mention of the theme x average negative confidence.
3 prioritised feature recommendations
Review-level results
Limitations: sentiment models can misread sarcasm, slang, Hinglish and domain terms; flagged rows should be checked by a human. Recommendations are decision support, not decisions.