Multi-hop Question Answering on 2WikiMultihopQA (EM, ASR)
0.6Exact Match (EM)Clean (No Attack)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Clean (No Attack)Framework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 0.6 | — | |
| IgnoreFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 0.59 | 0.01 | |
| IgnoreFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 0.59 | 0 | |
| Clean (No Attack)Framework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 0.58 | — | |
| CombinedFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 0.58 | 0 | |
| Clean (No Attack)Framework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 0.57 | — | |
| NaiveFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 0.56 | 0.03 | |
| NaiveFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 0.55 | 0 | |
| Clean (No Attack)Framework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 0.54 | — | |
| CombinedFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 0.53 | 0.03 | |
| PARADOXFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 0.51 | 0.03 | |
| NaiveFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 0.47 | 0.1 | |
| PoisonedRAGFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 0.41 | 0 | |
| PARADOXFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 0.41 | 0.15 | |
| PARADOXFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 0.38 | 0.1 | |
| NaiveFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 0.37 | 0.33 | |
| PoisonedRAGFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 0.37 | 0.02 | |
| PARADOXFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 0.36 | 0.21 | |
| PoisonedRAGFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 0.35 | 0.02 | |
| IgnoreFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 0.34 | 0.17 | |
| CombinedFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 0.34 | 0.01 | |
| PoisonedRAGFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 0.34 | 0.02 | |
| KidnapRAGFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 0.29 | 0.46 | |
| CombinedFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 0.26 | 0.37 | |
| IgnoreFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 0.24 | 0.18 | |
| KidnapRAGFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 0.2 | 0.39 | |
| KidnapRAGFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 0.09 | 0.54 | |
| KidnapRAGFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 0.09 | 0.62 |