Knowledge Graph construction on MuSiQue (Token Usage and Extraction Validity)
345Total TriplesIN-CONTEXT
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| IN-CONTEXTBackbone=Qwen3-30B-A3B2026.05 | 345 | 78.2 | 75 | 83.3 | 176.4 | 25.5 | 69.6 | |
| KGGENBackbone=Qwen3-30B-A3B2026.05 | 283 | — | 0 | — | 140.6 | 20.7 | 55.8 | |
| KGGENBackbone=GPT-OSS-120B2026.05 | 230 | — | 0 | — | 166.3 | 103.7 | 145.3 | |
| OAKBackbone=Qwen3-30B-A3B2026.05 | 224 | 58.7 | 89 | 29.3 | 65.5 | 14 | 30.4 | |
| OAK+MENDBackbone=Qwen3-30B-A3B2026.05 | 222 | 95.7 | 89 | 90.1 | 80 | 12.5 | 32.5 | |
| WIKONTICBackbone=Qwen3-30B-A3B2026.05 | 204 | 72.9 | 88 | — | 260.8 | 18.9 | 84.1 | |
| OAKBackbone=GPT-OSS-120B2026.05 | 195 | 63.3 | 85 | 32.2 | 66.6 | 54.8 | 71.4 | |
| OAK+MENDBackbone=GPT-OSS-120B2026.05 | 194 | 97.3 | 85 | 94.8 | 135.4 | 102.6 | 136.5 | |
| WIKONTICBackbone=GPT-OSS-120B2026.05 | 189 | 99.1 | 79 | — | 269.6 | 101.8 | 176 | |
| IN-CONTEXTBackbone=GPT-OSS-120B2026.05 | 112 | 97.3 | 18 | 98 | 181.9 | 39.5 | 85 |