GroupRAG: Cognitively Inspired Group-Aware Retrieval and Reasoning via Knowledge-Driven Problem Structuring

2026-03-26

AI Summary

Current language models often underperform in real-world scenarios due to insufficient knowledge and limited reasoning capabilities, compounded by an inability to effectively model the internal structure of complex problems. Inspired by cognitive science, this work proposes GroupRAG, a novel framework that introduces structured search mechanisms from human problem-solving into retrieval-augmented reasoning. GroupRAG explicitly identifies the problem’s group structure through knowledge-driven keypoint clustering and conducts parallel, structure-aware retrieval and reasoning from multiple conceptual starting points, enabling fine-grained coordination between retrieval and inference. Evaluated on the MedQA benchmark, GroupRAG significantly outperforms prevailing RAG and chain-of-thought approaches, demonstrating that structured problem modeling substantially enhances reasoning robustness.

Abstract

The performance of language models is commonly limited by insufficient knowledge and constrained reasoning. Prior approaches such as Retrieval-Augmented Generation (RAG) and Chain-of-Thought (CoT) address these issues by incorporating external knowledge or enforcing linear reasoning chains, but often degrade in real-world settings. Inspired by cognitive science, which characterizes human problem solving as search over structured problem spaces rather than single inference chains, we argue that inadequate awareness of problem structure is a key overlooked limitation. We propose GroupRAG, a cognitively inspired, group-aware retrieval and reasoning framework based on knowledge-driven keypoint grouping. GroupRAG identifies latent structural groups within a problem and performs retrieval and reasoning from multiple conceptual starting points, enabling fine-grained interaction between the two processes. Experiments on MedQA show that GroupRAG outperforms representative RAG- and CoT-based baselines. These results suggest that explicitly modeling problem structure, as inspired by human cognition, is a promising direction for robust retrieval-augmented reasoning.

Problem

• problem structure
• retrieval-augmented generation
• reasoning
• knowledge-driven
• cognitive science

Innovation

• GroupRAG
• knowledge-driven grouping
• problem structuring
• cognitive-inspired reasoning
• retrieval-augmented generation