reasoning-graphs

SciGraphSelect introduces a reference-grounded framework for auditable scientific data selection, organizing domain knowledge and reasoning operations into typed knowledge–reasoning graphs to score candidate documents with traceable evidence.

A Meta 与密歇根州立大学的新论文发现，LLM 构建的推理图质量差异很大，其错误会传导到基于这些图计算出的答案和置信度中；研究测试了 9 个模型，指出图的质量更多取决于模型家族与图结构而非规模。

The Large Knowledge Model (LKM) introduces a scientific knowledge infrastructure that transforms research papers into reasoning graphs, creating a Scientific Reasoning Landscape to support scientific search, question answering, and research planning with demonstrated accuracy improvements on benchmarks.

This paper introduces SeedRG, a semi-synthetic benchmark generation pipeline designed to eliminate knowledge leakage in Retrieval-Augmented Generation (RAG) evaluation by creating novel examples that preserve reasoning structures but are absent from model parametric memory.