evidence graph reasoning

To address fragmented evidence aggregation, insufficient modeling of reasoning paths, and limited interpretability in multi-hop fact verification, this paper proposes a structured reasoning framework. Methodologically, it explicitly models the dynamically evolving reasoning path as a subgraph structure spanning both retrieval and verification stages; introduces a synergistic dual-module architecture—structure-enhanced retrieval and path-guided verification; and incorporates structure-aware long-range dependency modeling. The approach integrates graph neural networks, incremental subgraph construction, and joint retrieval-verification optimization. Evaluated on FEVER and HoVer benchmarks, it significantly outperforms strong baselines, achieving simultaneous gains in verification accuracy and retrieval precision. Moreover, it improves reasoning consistency and enhances interpretability through transparent, graph-structured inference pathways.

This work addresses the semantic mismatch caused by structural heterogeneity in knowledge graph question answering and the lack of global structural awareness in existing methods. It reframes multi-hop reasoning as a schema-guided graph search task, constructing an adaptive query schema graph via a semantic-structure projection mechanism. By integrating a Triple-Dependent Graph Neural Network, the approach enables globally guided node anchoring and subgraph retrieval, thereby incorporating global structural information into the retrieval phase for the first time to generate high-quality evidence reasoning graphs. This strategy substantially improves both retrieval accuracy and evidence completeness of multi-hop reasoning paths, achieving state-of-the-art performance across multiple benchmark datasets.

This work addresses the limitations of current language model agents, which perform zero-shot reasoning for each query, resulting in low accuracy, high output variance, and an inability to reuse past reasoning. To overcome these issues, the authors propose a structured approach based on reasoning graphs and retrieval graphs that persistently stores evidence-level chains of thought as graph edges. They introduce a novel feedback mechanism centered on evidence relevance rather than query similarity, enabling new queries to trace back and reuse previously validated reasoning steps. This framework establishes a traceable, self-improving reasoning loop that significantly boosts accuracy and accelerates variance convergence on multi-hop question answering benchmarks. The entire decision process remains interpretable, and the method achieves these gains without fine-tuning the underlying language model.

Path matching in graph query languages (e.g., Cypher, SQL/PGQ, GQL) lacks a unified and efficient processing mechanism—particularly when supporting complex path semantics (e.g., shortest paths, simple paths) and regular-expression constraints on edge labels—posing dual challenges in expressive power and performance. This paper introduces the first cross-language, general-purpose path-solving framework. It features a compact symbolic path representation and integrates dynamic-programming-based enumeration, incremental pipelined execution, and regex compilation optimizations to enable unified modeling and efficient evaluation of diverse path semantics and edge-label constraints. Experimental evaluation on real-world datasets and complex queries demonstrates an order-of-magnitude speedup over state-of-the-art graph engines, while maintaining high expressiveness, strong scalability, and behavioral stability.

This work addresses the opacity of reasoning in large language model (LLM) agents performing data-intensive analysis, which hinders the verifiability of their conclusions. To resolve this, the authors propose VeriGraph—a traceable neuro-symbolic reasoning framework that constructs an explicit heterogeneous evidence directed acyclic graph (DAG) to unify raw data, variables, computational results, and natural language claims. VeriGraph introduces three evidence expansion primitives—computation, anchoring, and derivation—to enable structural traceability via graph reachability and incorporates claim-level evidence evaluation to quantify semantic support. Experimental results demonstrate that VeriGraph achieves state-of-the-art performance across four benchmarks, with its 8B variant attaining a claim-level anchoring accuracy of 87.61%, substantially enhancing the auditability and reproducibility of model outputs.

This work addresses the frequent failure of autonomous scientific agents due to unsupported claims or inconsistencies across research stages. To tackle this, the authors propose representing the agent’s internal state as a typed evidence graph comprising nodes for questions, knowledge gaps, hypotheses, experiments, findings, and claims. This framework enables, for the first time, real-time tracking of claim–evidence consistency, precise identification of flaws, and targeted repairs, with a graph checkpointing mechanism ensuring the safety of such interventions. The approach integrates evidence graph construction, consistency verification, regeneration of weak nodes, and chain-of-verification–guided paper generation. Evaluated on ARC-Bench-ML and NanoResearch-20, the method improves claim support rate by 40.19% and achieves 87.73% experimental data consistency, substantially outperforming baseline systems.