SGR: A Stepwise Reasoning Framework for LLMs with External Subgraph Generation
Xin Zhang, Yang Cao, Baoxing Wu, Kai Song, Siying Li
SGR builds query-specific subgraphs from external knowledge bases at inference time, then walks the LLM through multi-step reasoning grounded in that structured evidence rather than relying purely on parametric memory. The core idea is to decompose a complex question into intermediate steps, each anchored to relevant entities and relations extracted from the subgraph, then aggregate multiple reasoning paths into a final answer. Benchmark results show consistent gains over baselines on knowledge-intensive QA tasks, though margin sizes aren't specified in the abstract and the improvement appears incremental over existing KG-augmented reasoning methods.
No production traction yet. Zero citations on Semantic Scholar, and the GitHub references are all arxiv aggregator bots and personal paper-tracking pages โ none indicate actual implementation or downstream use. No code repository has been released.
Large Language Models (LLMs) have demonstrated strong capabilities across diverse NLP applications, such as translation, text generation, and question answering. Nevertheless, they remain limited in complex settings that demand deep reasoning and logical inference. Since these models are trained on large-scale text corpora, their generation process may still introduce irrelevant, noisy, or factually inconsistent content. To mitigate this problem, we introduce SGR, a stepwise framework that enhances LLM reasoning through external subgraph generation. SGR builds query-specific subgraphs from external knowledge bases and uses their semantic structure to support multi-step inference. By grounding intermediate reasoning steps in structured external knowledge, the framework helps the model concentrate on relevant entities, relations, and supporting evidence. In particular, SGR first constructs a subgraph tailored to the input question. It then guides the model to reason progressively over the generated structure and combines multiple reasoning trajectories to obtain the final prediction. Experimental results across several benchmark datasets show that SGR achieves consistent improvements over competitive baselines, highlighting its value for improving both reasoning accuracy and factual reliability.