Teaching-Feedback Classification - English Sentiment (Few-Shot): leaderboard
This benchmark re-runs a validated protocol for classifying open-ended teaching-evaluation comments, holding the annotation guide, splits and held-out evaluation fixed while varying the model. This board transfers the sentiment task to English course reviews whose three-class labels come from star ratings, with examples in the prompt. The score is weighted F1, and the prompt carries the label definitions from the annotation guide.
Metric: Weighted F1 (%; three-way sentiment classification (negative, neutral, positive) of open-ended teaching-evaluation comments, over a 1,500-comment cost-capped subset of the balanced 45,000-comment English corpus sampled from public RateMyProfessor reviews, whose three-class labels come from the star rating; few-shot prompting with the annotation guide's label definitions). Source: arxiv.org. Saturation forecast: Around February 2028. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Haiku 4.5 | 73.3 | #568 |
| 2 | Claude Opus 4.8 | 72.3 | #56 |
| 3 | Qwen 2.5 1.5B Instruct | 62 | #7470 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=teaching-feedback-classification-english-sentiment-few-shot · How It Works · Data refreshed daily, snapshot 2026-11-13.