Idea Generation on 50 seeds (test)
58Novelty (W)GoR-SFT
Evaluation Results
| Method | Links | |||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GoR-SFTOpponent=ResearchAgent, Backbone=7B base, Prompt format=same graph supervision, Evaluation protocol=fine-tuned2026.05 | 58 | 28 | 14 | 54 | 26 | 20 | 94 | 6 | 0 | 98 | 2 | 0 | 86 | 12 | 2 | 48 | 2 | 16.98 | 3.02 | |
| GoR-AgentOpponent=ResearchAgent, Backbone=gpt-4o, Prompt format=graph-format prompt, Evaluation protocol=zero-shot2026.05 | 38 | 46 | 16 | 44 | 28 | 28 | 92 | 4 | 4 | 96 | 2 | 2 | 84 | 10 | 6 | 47 | 3 | 15.88 | 4.12 | |
| GoR-SFTOpponent=CoI-Agent, Backbone=7B base, Prompt format=same graph supervision, Evaluation protocol=fine-tuned2026.05 | 32 | 24 | 44 | 24 | 28 | 48 | 78 | 16 | 6 | 76 | 16 | 8 | 70 | 16 | 14 | 40 | 10 | 13.26 | 6.74 | |
| GoR-AgentOpponent=CoI-Agent, Backbone=gpt-4o, Prompt format=graph-format prompt, Evaluation protocol=zero-shot2026.05 | 16 | 42 | 42 | 20 | 32 | 48 | 48 | 22 | 30 | 54 | 24 | 22 | 48 | 20 | 32 | 31 | 19 | 10.32 | 9.68 | |
| GoR-SFTOpponent=Si baseline, Backbone=7B base, Prompt format=same graph supervision, Evaluation protocol=fine-tuned2026.05 | 4 | 20 | 76 | 12 | 24 | 64 | 88 | 10 | 2 | 56 | 28 | 16 | 48 | 38 | 14 | 31 | 19 | 10.54 | 9.46 | |
| GoR-AgentOpponent=Si baseline, Backbone=gpt-4o, Prompt format=graph-format prompt, Evaluation protocol=zero-shot2026.05 | 0 | 26 | 74 | 6 | 30 | 64 | 76 | 12 | 12 | 38 | 34 | 28 | 34 | 36 | 30 | 26 | 24 | 9.08 | 10.92 |