Scientific Papers: Raw Full Text from S2ORC Collections

Approximately 57 million scientific papers with raw full text, extracted from multiple large-scale academic paper collections. The dataset is provided by scientifi-papers and was last updated on 2026-05-19. It includes subsets such as papers-2 (~18.5M papers), papers-3 (~27.4M papers), and pes2o (~8.2M papers).

Use Cases

• Pre-training large language models based on the raw full text of scientific papers.
• Fine-tuning specialized NLP models for tasks like scientific summarization or question-answering based on the full-text corpus.
• Building semantic search indices over scientific literature based on the extracted full text.

Strengths

• Contains approximately 57 million papers, indicating a large-scale corpus.
• Provides raw full text, suitable for deep text analysis.
• Comprises multiple subsets with defined sizes, such as papers-3 at ~27.4 million papers.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

S2ORC papers collection

Collection Method

Extracted from multiple large-scale academic paper collections.

Freshness

Last updated 2026-05-19 22:48:47; freshness should be verified.