CIRAG: Construction-Integration Retrieval and Adaptive Generation for Multi-hop Question Answering

2026-01-11

arXiv.org

AI Summary

This work addresses the limitations of existing iterative Retrieval-Augmented Generation (iRAG) methods in multi-hop question answering, which suffer from early error propagation due to greedy single-path retrieval and mismatched evidence granularity. To overcome these issues, the authors propose an iterative construct-and-integrate framework that preserves multiple plausible reasoning paths through triple-level query generation. They further design an adaptive cascaded multi-granularity mechanism that progressively expands evidence from triples to sentences and paragraphs, enabling fine-grained evidence fusion. Additionally, a teacher-student trajectory distillation strategy is introduced to enhance long-range reasoning efficiency. The proposed approach significantly outperforms current iRAG systems, achieving notable improvements in both answer accuracy and robustness to noisy evidence.

Technology Category

• Natural Language Processing: Question Answering
• Knowledge Representation and Reasoning: Argumentation
• Data Mining & Knowledge Management: Intelligent Query Processing

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• 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: Querying, indexing, and retrieval in Web-related graphs

Abstract

Triple-based Iterative Retrieval-Augmented Generation (iRAG) mitigates document-level noise for multi-hop question answering. However, existing methods still face limitations: (i) greedy single-path expansion, which propagates early errors and fails to capture parallel evidence from different reasoning branches, and (ii) granularity-demand mismatch, where a single evidence representation struggles to balance noise control with contextual sufficiency. In this paper, we propose the Construction-Integration Retrieval and Adaptive Generation model, CIRAG. It introduces an Iterative Construction-Integration module that constructs candidate triples and history-conditionally integrates them to distill core triples and generate the next-hop query. This module mitigates the greedy trap by preserving multiple plausible evidence chains. Besides, we propose an Adaptive Cascaded Multi-Granularity Generation module that progressively expands contextual evidence based on the problem requirements, from triples to supporting sentences and full passages. Moreover, we introduce Trajectory Distillation, which distills the teacher model's integration policy into a lightweight student, enabling efficient and reliable long-horizon reasoning. Extensive experiments demonstrate that CIRAG achieves superior performance compared to existing iRAG methods.

Problem

• multi-hop question answering
• iterative retrieval-augmented generation
• evidence granularity
• reasoning paths
• noise control

Innovation

• Iterative Construction-Integration
• Adaptive Cascaded Multi-Granularity Generation
• Trajectory Distillation
• Multi-hop Question Answering
• Retrieval-Augmented Generation