arXiv cs.AIMonday · May 25, 2026FREE

SciAtlas: A Large-Scale Knowledge Graph for Automated Scientific Research

SciAtlas, introduced in a new arXiv preprint (2605.22878v1), is a large-scale, multi-disciplinary academic knowledge graph designed to address the fragmentation of scientific knowledge. It integrates over 43 million papers from 26 disciplines, totaling 157 million entities and 3 billion triplets, forming a panoramic scientific evolution network. The graph serves as a structured topological cognitive substrate that dismantles disciplinary barriers and provides AI agents with a global perspective. To leverage this graph, the authors developed a neuro-symbolic retrieval algorithm featuring tri-path collaborative recall and graph reranking. This approach aims to overcome the limitations of current academic retrieval tools that rely on superficial keyword matching or vector-space semantic retrieval, which lack topological reasoning capabilities. The algorithm is designed to reduce logical hallucinations and high inference costs often associated with agentic deep-research frameworks. The paper is available on arXiv and represents a significant step toward automated scientific research by enabling more accurate and efficient knowledge discovery across disciplines.