pyEuropePMC

arXiv client

ArxivClient searches arXiv preprints through the arXiv API and returns LiteratureResult records. The API needs no key or registration.

Search

from pyeuropepmc.features.search import ArxivClient

with ArxivClient() as client:
    papers = client.search("machine learning", limit=10)

for paper in papers:
    print(paper.source_id, paper.publication_year, paper.title)

search(query, limit=25, sort=None, **kwargs) -> list[LiteratureResult]

Parameter	Type	Default	Description
`query`	`str`	required	arXiv query, sent unchanged as `search_query`
`limit`	`int`	`25`	Maximum records; capped at 2000
`sort`	`str` or `None`	`None`	`"relevance"`, `"date"` (submission date) or `"title"`; `None` leaves the order to arXiv
`categories`	`str` or `list[str]`	not sent	arXiv subject categories to restrict the search to, as a list or a comma-separated string
`id_list`	`str`	not sent	Comma-separated arXiv IDs to look up

Other keyword arguments are ignored without a warning.

Fields and subject categories

Queries use arXiv’s own syntax: prefixes such as ti: (title), au: (author), abs: (abstract), cat: (subject category) and all:, combined with AND, OR and ANDNOT. To restrict a search to subject categories, put cat: terms in the query:

from pyeuropepmc.features.search import ArxivClient

with ArxivClient() as client:
    papers = client.search(
        'all:"quantum computing" AND (cat:cs.AI OR cat:cs.LG)',
        limit=20,
        sort="date",
    )

print(len(papers))

categories builds those cat: clauses for you. The categories are OR-ed together and AND-ed onto the query, so

client.search("quantum computing", categories=["cs.AI", "cs.LG"])

sends (quantum computing) AND (cat:cs.AI OR cat:cs.LG). A comma-separated string works too: categories="cs.AI, cs.LG".

The query syntax is described in the arXiv API user manual. Through UnifiedSearch, a query without arXiv prefixes is sent as all:"<query>".

Malformed requests

arXiv reports a rejected request as an ordinary Atom feed holding one entry titled Error. The client recognises those entries, logs the message at warning level and returns an empty list, so an error never arrives as a search result.

Get a paper

from pyeuropepmc.features.search import ArxivClient

with ArxivClient() as client:
    by_id = client.get_paper("2101.12345")
    by_doi = client.get_paper("10.48550/arXiv.2101.12345")
    by_url = client.get_paper("https://arxiv.org/abs/2101.12345")

print(by_id.title if by_id else "not found")

get_paper(identifier) -> LiteratureResult | None accepts an arXiv ID, an ID with an arxiv: prefix, an abstract URL or an arXiv DOI (10.48550/arXiv.<id>).

Result fields

Field	Content
`source`	`"arxiv"`
`source_id`	arXiv ID without the version suffix, for example `2101.12345`
`title`, `abstract`	Entry title and summary
`authors`	`Author` objects with names in `Last, First` form
`publication_year`	Year of the first version
`doi`	DOI of the published version, when the authors added one
`journal`	Journal reference, when present

Subject categories are not included in the results.

Rate limit

ArxivClient(rate_limit_delay=3.0, timeout=30, cache_config=None) waits 3 seconds between requests, following arXiv’s request to send at most one request every three seconds. UnifiedSearch replaces this with its own rate_limit_delay, 1.2 seconds by default; use UnifiedSearch(rate_limit_delay=3.0) when arXiv is one of the sources.