Optimizing Question Semantic Space for Dynamic Retrieval-Augmented Multi-hop Question Answering

2025-05-31

AI Summary

In multi-hop question answering, conventional single-step retrieve-and-read paradigms suffer from semantic mismatch and inefficient, inaccurate retrieval due to strong inter-dependencies among sub-questions. This paper proposes Q-DREAM, a dynamic retrieval-augmented framework: it first structurally decomposes complex questions into interrelated sub-questions; then models their dependencies via a graph neural network and refines semantic embeddings through contrastive learning. Crucially, it introduces a sub-question dependency-aware dynamic retrieval mechanism that enables stepwise, semantically aligned retrieval. Evaluated on multiple benchmarks, Q-DREAM achieves state-of-the-art accuracy, with significant cross-domain generalization gains. It improves retrieval efficiency by 32% over recent baselines while reducing redundant computation. The framework thus delivers high accuracy, strong domain transferability, and computational efficiency—addressing key limitations of static, monolithic retrieval approaches in multi-hop QA.

Technology Category

• Natural Language Processing: Question Answering
• Data Mining & Knowledge Management: Intelligent Query Processing
• Computer Vision: Multi-modal Vision

Application Category

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

Abstract

Retrieval-augmented generation (RAG) is usually integrated into large language models (LLMs) to mitigate hallucinations and knowledge obsolescence. Whereas,conventional one-step retrieve-and-read methods are insufficient for multi-hop question answering, facing challenges of retrieval semantic mismatching and the high cost in handling interdependent subquestions. In this paper, we propose Optimizing Question Semantic Space for Dynamic Retrieval-Augmented Multi-hop Question Answering (Q-DREAM). Q-DREAM consists of three key modules: (1) the Question Decomposition Module (QDM), which decomposes multi-hop questions into fine-grained subquestions; (2) the Subquestion Dependency Optimizer Module (SDOM), which models the interdependent relations of subquestions for better understanding; and (3) the Dynamic Passage Retrieval Module (DPRM), which aligns subquestions with relevant passages by optimizing the semantic embeddings. Experimental results across various benchmarks demonstrate that Q-DREAM significantly outperforms existing RAG methods, achieving state-of-the-art performance in both in-domain and out-of-domain settings. Notably, Q-DREAM also improves retrieval efficiency while maintaining high accuracy compared with recent baselines.

Problem

• Optimizing semantic space for multi-hop QA retrieval
• Reducing retrieval mismatching in interdependent subquestions
• Improving efficiency and accuracy in dynamic passage retrieval

Innovation

• Decomposes multi-hop questions into subquestions
• Models subquestion dependencies for better understanding
• Optimizes semantic embeddings for dynamic retrieval