DualResearch: Entropy-Gated Dual-Graph Retrieval for Answer Reconstruction

2025-10-09

AI Summary

To address context contamination, weak evidential support, and fragile execution paths in multi-tool collaborative scientific reasoning, this paper proposes a dual-graph retrieval framework integrated with entropy-controlled fusion. The framework constructs a semantic breadth graph—modeling tool coverage—and a causal depth graph—capturing causal interpretability of reasoning chains. An entropy-controlled gating mechanism dynamically weights heterogeneous evidence in log-space, enabling reliability-driven path selection and consistency enhancement. Additionally, we introduce seed-anchored semantic diffusion, causal path matching, and layer-native relevance functions, augmented by global calibration. Evaluated on the HLE and GPQA benchmarks, our method achieves substantial improvements: +7.7% accuracy on HLE and +6.06% on GPQA, demonstrating robustness and effectiveness for complex, multi-step scientific reasoning.

Technology Category

• Reasoning under Uncertainty: Causality
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

The deep-research framework orchestrates external tools to perform complex, multi-step scientific reasoning that exceeds the native limits of a single large language model. However, it still suffers from context pollution, weak evidentiary support, and brittle execution paths. To address these issues, we propose DualResearch, a retrieval and fusion framework that matches the epistemic structure of tool-intensive reasoning by jointly modeling two complementary graphs: a breadth semantic graph that encodes stable background knowledge, and a depth causal graph that captures execution provenance. Each graph has a layer-native relevance function, seed-anchored semantic diffusion for breadth, and causal-semantic path matching with reliability weighting for depth. To reconcile their heterogeneity and query-dependent uncertainty, DualResearch converts per-layer path evidence into answer distributions and fuses them in log space via an entropy-gated rule with global calibration. The fusion up-weights the more certain channel and amplifies agreement. As a complement to deep-research systems, DualResearch compresses lengthy multi-tool execution logs into a concise reasoning graph, and we show that it can reconstruct answers stably and effectively. On the scientific reasoning benchmarks HLE and GPQA, DualResearch achieves competitive performance. Using log files from the open-source system InternAgent, its accuracy improves by 7.7% on HLE and 6.06% on GPQA.

Problem

• Addresses context pollution in multi-step scientific reasoning
• Improves evidentiary support through dual-graph retrieval framework
• Stabilizes execution paths via entropy-gated answer reconstruction

Innovation

• Models two complementary graphs for tool reasoning
• Uses entropy-gated fusion to reconcile graph heterogeneity
• Compresses execution logs into concise reasoning graphs