Reading Comprehension Item Generation on Difficulty-controllable RC Dataset
52.26DAS ScoreMAFIGQwen3−32B
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| MAFIGQwen3−32BDifficulty Control Granularity=Feature-based2026.05 | 52.26 | 92.29 | 99.32 | 2.9242 | 95.18 | 94.29 | |
| DirectGPT−5Difficulty Control Granularity=Feature-based2026.05 | 49.52 | 2.5 | 77.81 | 2.9105 | 90.94 | 92.41 | |
| DirectGPT−5Difficulty Control Granularity=Level-based2026.05 | 29.49 | — | — | 2.9816 | 93.32 | 94.08 | |
| DirectQwen3−32BDifficulty Control Granularity=Feature-based2026.05 | 27.59 | 0 | 59.1 | 2.6094 | 93.68 | 93.93 | |
| IncrementalGPT−5Difficulty Control Granularity=Level-based2026.05 | 27.5 | — | — | 2.9637 | 93.48 | 93.09 | |
| IncrementalQwen3−32BDifficulty Control Granularity=Level-based2026.05 | 18.04 | — | — | 2.5605 | 93.32 | 94.08 | |
| DirectQwen3−32BDifficulty Control Granularity=Level-based2026.05 | 10.37 | — | — | 2.6371 | 93.55 | 92.8 |