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

TL;DR

This work tackles the challenge of multi-hop question answering by mitigating retrieval noise and addressing granularity mismatch. It introduces CIRAG, a two-module framework with Iterative Construction-Integration (ICI) to preserve multiple evidence paths and synthesize next-hop queries, and Adaptive Cascaded Multi-Granularity Knowledge-Augmented Generation (ACMG) to progressively expand context from triples to sentences and passages. To enable practical long-horizon reasoning, Trajectory Distillation transfers the integration policy from a strong teacher to a lightweight student. Empirical results on HotpotQA, 2WikiMultiHopQA, and MuSiQue show CIRAG consistently outperforms baselines, with evidence that most questions are solvable using triples and that adaptive cascades effectively balance noise control with contextual sufficiency. The approach also generalizes to WebQA and NQ, highlighting its robustness across QA tasks and model scales.

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.

Figures (12)

• Figure 1: Challenges in Triple-based Retrieval.
• Figure 2: Overview of CIRAG. At each iteration, it employs an Iterative Construction-Integration (ICI) Retrieval module to retrieve a core triple set and record provenance links to their supporting sentences and documents, including two iterative phases: construction phase and integration phase. The core triple set is used to produce the final answer via the Adaptive Cascaded Multi-Granularity Knowledge-Enhanced Generation (ACMG) module.
• Figure 3: Context granularity distribution across datasets
• Figure 4: Comparison of single-granularity and cascaded performance
• Figure 5: Effect of Trajectory Distillation on context granularity distribution