A Multi-Agent Approach for Claim Verification from Tabular Data Documents

Benchmark Model Rank Results
table-based-fact-verification-on-sem-tab-factsMACE (LLaMA-8B, Pm+Em+Vm')–Micro F1 (Subtask A): 86.0
table-based-fact-verification-on-sem-tab-factsMACE (Mistral-7B, Pm+Em+Vm')–Micro F1 (Subtask A): 74.0
table-based-fact-verification-on-sem-tab-factsMACE (Qwen-235B, Pm+Em)–Micro F1 (Subtask A): 90.0
table-based-fact-verification-on-sem-tab-factsMACE (Qwen-72B, Pm+Em+Vm')–Micro F1 (Subtask A): 89.0
table-claim-verification-on-scitabMACE (LLaMA-8B, Pm + Em + Vm)–Macro-F1 (3-class): 0.6500
table-claim-verification-on-scitabMACE (Mistral-7B, Pm + Em + Vm)–Macro-F1 (3-class): 0.5800
table-claim-verification-on-scitabMACE (Qwen-235B, Pm + Em + Vm)–Macro-F1 (3-class): 0.7100
table-claim-verification-on-scitabMACE (Qwen-72B, Pm + Em + Vm)–Macro-F1 (3-class): 0.6700
table-claim-verification-on-scitabQwen-235B + CoT (baseline)–Macro-F1 (3-class): 0.7300