Scholar-kg: 1 Million Papers as a Queryable Knowledge Graph

Approximately 1 million academic papers from sources like arXiv and bioRxiv have been processed into a unified, multi-layered knowledge graph. The dataset, created by InternScience, decomposes each paper into five modules covering metadata, entities, abstracted knowledge, citation context, and fine-grained relations. It was last updated on June 12, 2026.

Use Cases

• Building a semantic search engine for academic papers based on the decomposed knowledge modules.
• Training or evaluating information extraction models on the structured entities and relations.
• Analyzing citation networks and research trends using the provided citation context.
• Developing question-answering agents for scientific domains based on the abstracted knowledge representations.

Strengths

• Contains about 1 million papers, providing a substantial scale for analysis.
• Papers are decomposed into a structured, five-module representation (A–E) designed for querying.
• Sources include established preprint repositories like arXiv and bioRxiv.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The description references a full page for details, indicating core metadata may be incomplete here.

Provenance

Source

Processed from arXiv, bioRxiv, and other unspecified sources by Agents-K1.

Collection Method

Papers were decomposed into a unified, queryable representation organized into five modules.

Freshness

Last updated 2026-06-12 03:03:40; freshness should be verified.