paperscraper

paperscraper is a python package for scraping publication metadata or full text files (PDF or XML) from PubMed or preprint servers such as arXiv, medRxiv, bioRxiv, and chemRxiv. It provides a streamlined interface to scrape metadata, retrieve citation counts from Google Scholar or Semantic Scholar, query journal impact factors, and run simple postprocessing and plotting routines for meta-analysis.

Table of Contents

1. Getting Started
   • Download xRxiv Dumps
   • arXiv Local Dump
2. Examples
   • Paper Keyword Analysis
   • PDF Retrieval
   • Scholar Metrics Analysis
   • Self-Citation Analysis
3. Citation
4. Contributors

Getting started

pip install paperscraper

or, with uv:

uv add paperscraper

This is enough to query PubMed, arXiv or Google Scholar.

Local development

uv sync

This installs the project and dev tooling into .venv. Use uv run to execute commands, for example:

uv run python -c "import paperscraper"

Download xRxiv Dumps

However, to scrape publication data from the preprint servers bioRxiv, medRxiv and chemRxiv, the setup is different. The entire history of papers is downloaded and stored in the server_dumps folder in JSONL format (one paper per line).

from paperscraper.get_dumps import biorxiv, medrxiv, chemrxiv
chemrxiv()  #  Takes <15min -> +50K papers (~30 MB file)
medrxiv()  #  Takes <5min -> +100K papers (~200 MB file)
biorxiv()  # Takes <1h -> +450K papers (~800 MB file)

NOTE: Once the dumps are stored, please make sure to restart the python interpreter so that the changes take effect. NOTE: If you experience API connection issues, retries and request behavior can be tuned, e.g.:

biorxiv(
    max_retries=12,
    request_timeout=(5.0, 45.0),      # connect timeout, read timeout
    retry_backoff_seconds=1.0,        # initial retry backoff
    max_workers=8,                    # number of parallel date windows
    window_days=30,                   # smaller windows increase parallelism
)

paperscraper also allows scraping {med/bio/chem}rxiv for specific dates.

medrxiv(start_date="2023-04-01", end_date="2023-04-08")

But watch out. The resulting .jsonl file will be labelled according to the current date and all your subsequent searches will be based on this file only. If you use this option you might want to keep an eye on the source files (paperscraper/server_dumps/*jsonl) to ensure they contain the paper metadata for all papers you're interested in. Use paperscraper.utils.get_server_dumps_dir() to inspect the active dump directory.

arXiv local dump

Local search can be faster than using the arXiv API, especially if you plan many queries. Paperscraper provides two backends to bulk-download arXiv, Kaggle and the arxiv package. The default is kaggle since it is much faster. Before using it, authenticate with your Kaggle account:

kaggle auth login

from paperscraper.get_dumps import arxiv
arxiv(start_date='2019-01-01', end_date='2026-12-31')

NOTE: The disadvantage of the kaggle backend is that it bulk-downloads all of arXiv. For small API-backed dumps, better use the arxiv PyPI package backend:

from paperscraper.get_dumps import arxiv
arxiv(start_date='2024-01-01',end_date='2024-01-04',backend='api')

Afterwards you can search the local arXiv dump just like the other x-rxiv dumps. The direct endpoint is paperscraper.arxiv.get_arxiv_papers_local. You can also specify the backend directly in the get_and_dump_arxiv_papers function:

from paperscraper.arxiv import get_and_dump_arxiv_papers
get_and_dump_arxiv_papers(..., backend='local')

Examples

paperscraper is built on top of arxiv, pymed, and scholarly.

The README keeps examples short. The documentation site has fuller walkthroughs:

• Paper Keyword Analysis
• PDF Retrieval
• Scholar Metrics Analysis
• Self-Citation Analysis

Paper keyword analysis

Nested keyword lists encode Boolean logic: outer lists are AND, inner lists are synonyms with OR.

from paperscraper.pubmed import get_and_dump_pubmed_papers

ai = ["Artificial intelligence", "Machine learning"]
qc = ["Quantum computing", "Quantum information", "Quantum algorithm"]
chemistry = ["Chemistry", "Chemical", "Molecule", "Materials science"]

get_and_dump_pubmed_papers([ai, qc, chemistry], "ai_quantum_chemistry.jsonl")

This writes matching PubMed records to ai_quantum_chemistry.jsonl.

For local bioRxiv, medRxiv, or chemRxiv search, download the dumps once as described in Download xRxiv Dumps, restart Python, then use dump_queries to query all available backends:

from paperscraper import dump_queries

dump_queries([[ai, qc, chemistry]], ".")

See the paper keyword analysis example for arXiv, Google Scholar, multi-database querying, and plotting.

PDF Retrieval

Download a PDF or XML by DOI:

from paperscraper.pdf import save_pdf

save_pdf({"doi": "10.48550/arXiv.2207.03928"}, filepath="gt4sd_paper.pdf")

Output: True when the file was saved.

See the PDF retrieval example for batch downloads, fallbacks, publisher API keys, and downstream PDF analysis.

Scholar metrics analysis

Get paper citation counts, Google Scholar search results, and journal metrics:

from paperscraper.citations import (
    get_bibtex_entry,
    get_citations_by_doi,
    get_citations_from_title,
    get_citing_papers_from_title,
    get_endnote_entry,
)
from paperscraper.impact import Impactor
from paperscraper.scholar import get_scholar_author_papers, get_scholar_papers

get_citations_by_doi("10.1021/acs.jcim.3c00132")
get_citations_from_title(
    "GT4SD: Generative Toolkit for Scientific Discovery",
    backend="searchapi",
)
get_scholar_papers(
    "GT4SD",
    backend="searchapi",
    search_api_kwargs={"top_k": 5, "num_enrich": 1},
)
get_scholar_author_papers("Jannis Born", max_results=5)

title = "Quantum doubly stochastic transformers"
get_bibtex_entry(title)
get_endnote_entry(title)
get_citing_papers_from_title(title)
get_citing_papers_from_title(title, max_results=5, full_info=True)
Impactor().search("Nat Comms", threshold=85, sort_by="impact")

Set SEARCH_API_KEY to use the SearchAPI Google Scholar backend. Outputs include citation counts and Scholar search metadata (author list, title, abstract, journal). Citing-paper results are Paper objects; full_info=True also resolves available DOIs and authors. Author lookup prefers an exact Scholar profile and falls back to an author:"name" search when no profile exists; its default limit is 30 papers.

Author-level Semantic Scholar metrics can be retrieved by Semantic Scholar ID, name, or ORCID:

from paperscraper.citations.orcid import orcid_to_author_name
from paperscraper.citations.utils import author_name_to_ssaid, semantic_scholar_requests_get

ssaid, name = author_name_to_ssaid(orcid_to_author_name("0000-0001-8307-5670"))
metrics = semantic_scholar_requests_get(
    f"https://api.semanticscholar.org/graph/v1/author/{ssaid}",
    params={"fields": "paperCount,citationCount,hIndex"},
).json()

Output: a JSON object with paperCount, citationCount, and hIndex (for example, 63, 1910, and 21 for the ORCID above).

See the scholar metrics analysis example for Google Scholar, Semantic Scholar, researcher metrics, and journal impact factors.

Self-citation analysis

Estimate paper-level self-citations and self-references:

from paperscraper.citations import self_citations_paper, self_references_paper

doi = "10.1038/s41586-023-06600-9"
self_citations_paper(doi).citation_score
self_references_paper(doi).reference_score

Output: 3.192 and 5.05, the mean self-citation and self-reference percentages across paper authors.

The documentation example also includes a small researcher-level benchmark with self-citation and self-reference trends by discipline, career-stage group, and an overall average across disciplines:

See the self-citation analysis example for paper- and author-level workflows using Semantic Scholar.

Citation

If you scrape papers with paperscraper, please cite the paperscraper paper :)

@article{born2021trends,
  title={Trends in Deep Learning for Property-driven Drug Design},
  author={Born, Jannis and Manica, Matteo},
  journal={Current Medicinal Chemistry},
  volume={28},
  number={38},
  pages={7862--7886},
  year={2021},
  publisher={Bentham Science Publishers}
}

Contributing and support

Contribution guidelines are in CONTRIBUTING.md, support expectations are in SUPPORT.md, and project decision-making is summarized in GOVERNANCE.md. Release notes are maintained with GitHub releases and PyPI release history rather than in a separate changelog.

Contributors

Thanks to the following contributors:

• @mathinic: improved PubMed full text retrieval with additional fallback mechanisms (BioC-PMC, eLife and optional Wiley/Elsevier APIs).
• @memray: added automatic retries when downloading the {med/bio/chem}rxiv dumps.
• @achouhan93: added date-bounded scraping for {med/bio/chem}rxiv.
• @daenuprobst: added direct PDF scraping via paperscraper.pdf.save_pdf.
• @oppih: added chemRxiv DOI and URL metadata where available.
• @lukasschwab: enabled support for arxiv > 1.4.2.
• @juliusbierk: bug fixes.