Scientific Reasoning Graph Extraction and Introduction Generation on Scientific Papers (test)
48.6Precision/Correlation ScoreLECTOR
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LECTORBackbone=Qwen3-4B2026.05 | 48.6 | 74.5 | 62.3 | 83.4 | 53 | 66.5 | |
| One-Step-BaselineBackbone=Qwen3-4B2026.05 | 47.6 | — | — | 82.9 | 47.7 | — | |
| Claude-haiku-4.52026.05 | 47 | 70.7 | 72.7 | 69.9 | 52.9 | 61.2 | |
| Grok42026.05 | 46.4 | 65.1 | 60.1 | 69.1 | 72.1 | 59.9 | |
| Gemini-2.5pro2026.05 | 45.8 | 35.7 | 26.2 | 81.4 | 71 | 56.6 | |
| Qwen3-4B-Instruct-2507 (Base)Backbone=Qwen3-4B2026.05 | 45.3 | 47.8 | 68.2 | 54.6 | 44.4 | 51 | |
| GLM-4.72026.05 | 42.8 | 12.3 | 8.8 | 73.4 | 73.9 | 46.6 | |
| GPT-o32026.05 | 41.6 | 69 | 44.8 | 88.2 | 69.1 | 65.6 |