Knowledge Insertion on CrossCult-KIBench single-insert
91.12Reliability (ROUGE-L)FineTune
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| FineTuneBase Model=InternVL3.5-8B2026.05 | 91.12 | 9.35 | 24.66 | 6.32 | 5.36 | 2.28 | 4.96 | 0.9 | 31.53 | 4.71 | |
| MCKI (Ours)Base Model=InternVL3.5-8B2026.05 | 79.83 | 8.86 | 22.2 | 5.71 | 100 | 9.99 | 89.66 | 9.47 | 72.92 | 8.51 | |
| MSCKEBase Model=Qwen3.5-9B2026.05 | 71.27 | 8.5 | 27.58 | 6.27 | 100 | 10 | 28.64 | 4.94 | 56.87 | 7.43 | |
| IKEBase Model=InternVL3.5-8B2026.05 | 70.25 | 8.54 | 23.58 | 6.28 | 24.92 | 5.14 | 34.46 | 6.45 | 38.3 | 6.6 | |
| MSCKEBase Model=InternVL3.5-8B2026.05 | 69.93 | 8.48 | 23.55 | 5.8 | 100 | 9.99 | 91.55 | 9.38 | 71.26 | 8.41 | |
| SERACBase Model=Qwen3.5-9B2026.05 | 67.59 | 8.23 | 27.59 | 6.42 | 5.75 | 3.98 | 18.88 | 3.92 | 29.95 | 5.64 | |
| SERACBase Model=InternVL3.5-8B2026.05 | 57.69 | 7.64 | 22.56 | 5.63 | 78.81 | 9.07 | 26.51 | 4.38 | 46.39 | 6.68 | |
| FineTuneBase Model=Qwen3.5-9B2026.05 | 55.9 | 6.35 | 15.54 | 4.37 | 0 | 0.96 | 3.17 | 0.61 | 18.65 | 3.07 | |
| MCKI (Ours)Base Model=Qwen3.5-9B2026.05 | 50.66 | 7.77 | 24.43 | 6.74 | 100 | 10 | 88.27 | 9.32 | 65.84 | 8.46 | |
| IKEBase Model=Qwen3.5-9B2026.05 | 47.25 | 8.06 | 23.76 | 6.84 | 19.93 | 5.2 | 35.72 | 5.63 | 31.66 | 6.43 | |
| MENDBase Model=Qwen3.5-9B2026.05 | 44.91 | 6.52 | 26.31 | 5.8 | 81.83 | 9.22 | 61.32 | 8.03 | 53.59 | 7.39 | |
| MENDBase Model=InternVL3.5-8B2026.05 | 19.24 | 4.82 | 19.33 | 4.87 | 67.16 | 8.58 | 42.59 | 6.89 | 37.08 | 6.29 | |
| Base ModelBase Model=Qwen3.5-9B2026.05 | 14.13 | 4.5 | 14.06 | 4.43 | 100 | 10 | 100 | 10 | 57.05 | 7.23 | |
| Base ModelBase Model=InternVL3.5-8B2026.05 | 12.24 | 4.02 | 12.23 | 4.03 | 100 | 9.99 | 100 | 9.98 | 56.12 | 7 |