Intern-Atlas: AI Method Evolution Graph from 1M+ Papers
A team of researchers has introduced Intern-Atlas, a graph that represents the evolution of methodologies by automatically detecting method-level entities and deducing lineage connections among AI research methods. This graph is constructed from a vast collection of 1,030,314 papers, including those from AI conferences, journals, and arXiv preprints, and features 9,410,201 semantically categorized nodes. It highlights the obstacles that influence the progression of innovations, addressing the shortcomings of current document-focused research systems that fail to provide clear representations of methodological evolution. This new infrastructure aims to assist AI-driven research agents in effectively reconstructing method evolution topologies from unstructured text.
Key facts
- Intern-Atlas is a methodological evolution graph for AI research.
- It automatically identifies method-level entities and infers lineage relationships.
- The graph is built from 1,030,314 papers.
- Sources include AI conferences, journals, and arXiv preprints.
- The graph comprises 9,410,201 semantically typed nodes.
- It captures bottlenecks that drive transitions between innovations.
- Existing research infrastructure is document-centric and lacks method evolution representations.
- The tool targets AI-driven research agents as consumers of scientific knowledge.
Entities
Institutions
- arXiv