Hallucination Mitigation in Recursive Multi-agent Reasoning on TruthfulQA and TriviaQA
86.7AccuracyProposed HPR-adaptive
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Proposed HPR-adaptiveAgent Architecture=Recursive Multi-agent, Defense Strategy=HPR-adaptive2026.06 | 86.7 | 7.2 | 83.6 | 1.08 | 92.8 | 0.81 | |
| KnowHaluAgent Architecture=Recursive Multi-agent, Defense Strategy=KnowHalu2026.06 | 84.1 | 8.9 | 81.4 | 1.17 | 90.9 | 0.92 | |
| LapEigValsAgent Architecture=Recursive Multi-agent, Defense Strategy=LapEigVals2026.06 | 83.6 | 9.3 | 81 | 1.19 | 90.5 | 0.93 | |
| Entropy-based detectionAgent Architecture=Recursive Multi-agent, Defense Strategy=Entropy-based detection2026.06 | 83.2 | 9.7 | 80.6 | 1.21 | 90.1 | 0.95 | |
| SelfCheckGPTAgent Architecture=Recursive Multi-agent, Defense Strategy=SelfCheckGPT2026.06 | 82.6 | 10.1 | 80.1 | 1.24 | 89.7 | 0.97 | |
| Multi-agent, no defenseAgent Architecture=Recursive Multi-agent, Defense Strategy=None2026.06 | 81.2 | 11.8 | 78.4 | 1.34 | 88.2 | 1.08 | |
| Single-agent LLMAgent Architecture=Single-agent2026.06 | 79.1 | 13.1 | 74.8 | 1 | 86.9 | 0.34 |