ToE: A Hierarchical and Explainable Claim Verification Framework with Dynamic Multi-source Evidence Retrieval and Aggregation

2026-06-26

AI Summary

This work addresses the spread of misinformation caused by retrieval-augmented large language models (LLMs) under Generative Engine Optimization (GEO) poisoning attacks. To counter this, the authors propose the Tree of Evidence (ToE) framework, which models claims as dynamically expanding argumentation trees. ToE iteratively constructs interpretable evidence chains through reinforcement learning–driven multi-source retrieval, evidence credibility assessment, and hierarchical aggregation. The framework innovatively employs a dynamic argumentation tree structure to enable layered, interpretable fact-checking and provides theoretical guarantees by deriving an upper bound on retrieval error, ensuring policy convergence to an information-theoretically near-optimal neighborhood. Experiments demonstrate that ToE significantly outperforms strong baselines across multiple datasets and LLM backbones, achieving accuracy gains of 4–24 percentage points, with particularly robust performance in adversarial poisoning scenarios.

Abstract

The rapid spread of fake news poses increasing threats to information ecosystems, especially as AI-generated misinformation under Generative Engine Optimization (GEO) poisoning allows adversarially crafted content to be systematically surfaced by retrieval systems, contaminating LLM reasoning. In this paper, we propose Tree of Evidence (ToE), a hierarchical evidence reasoning framework for automated fact-checking that models each claim as a dynamically expanding argument tree. ToE integrates a reinforcement learning-driven multi-source retrieval agent, an evidence evaluation agent, and an argument tree aggregation algorithm to iteratively decompose, retrieve, and verify claims through an explainable evidence chain. We further provide a theoretical analysis of the retrieval process, deriving a formal error bound that guarantees the learned policy converges to a neighborhood of the information-theoretically optimal policy. Experiments across multiple datasets and backbone LLMs demonstrate that ToE achieves improvements ranging from 4 to 24 percentage points over competitive baselines, with particularly pronounced gains on adversarially poisoned inputs.

Problem

• fake news
• Generative Engine Optimization
• evidence retrieval
• fact-checking
• misinformation

Innovation

• Tree of Evidence
• Dynamic Multi-source Retrieval
• Explainable Claim Verification
• Reinforcement Learning Agent
• Adversarial Robustness