Skip to main content

paperscraper: Package to scrape papers.

Project description

build License: MIT PyPI version Downloads Downloads Code style: black

paperscraper

Overview

paperscraper is a python package that ships via pypi and facilitates scraping publication metadata from PubMed or from preprint servers such as arXiv, medRxiv, bioRxiv and chemRxiv. It provides a streamlined interface to scrape metadata and comes with simple postprocessing functions and plotting routines for meta-analysis.

Getting started

pip install paperscraper

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

Download X-rxiv Dumps

However, to scrape publication data from the preprint servers biorxiv, medrxiv and chemrxiv, the setup is different. The entire dump is downloaded and stored in the server_dumps folder in a .jsonl format (one paper per line).

from paperscraper.get_dumps import biorxiv, medrxiv, chemrxiv
medrxiv()  #  Takes ~30min and should result in ~35 MB file
biorxiv()  # Takes ~1h and should result in ~350 MB file
chemrxiv()  #  Takes ~45min and should result in ~20 MB file

NOTE: Once the dumps are stored, please make sure to restart the python interpreter so that the changes take effect.

Examples

paperscraper is build on top of the packages pymed, arxiv and scholarly.

Publication keyword search

Consider you want to perform a publication keyword search with the query: COVID-19 AND Artificial Intelligence AND Medical Imaging.

  • Scrape papers from PubMed:
from paperscraper.pubmed import get_and_dump_pubmed_papers
covid19 = ['COVID-19', 'SARS-CoV-2']
ai = ['Artificial intelligence', 'Deep learning', 'Machine learning']
mi = ['Medical imaging']
query = [covid19, ai, mi]

get_and_dump_pubmed_papers(query, output_filepath='covid19_ai_imaging.jsonl')
  • Scrape papers from arXiv:
from paperscraper.arxiv import get_and_dump_arxiv_papers

get_and_dump_arxiv_papers(query, output_filepath='covid19_ai_imaging.jsonl')
  • Scrape papers from bioRiv, medRxiv or chemRxiv:
from paperscraper.xrxiv.xrxiv_query import XRXivQuery

querier = XRXivQuery('server_dumps/chemrxiv_2020-11-10.jsonl')
querier.search_keywords(query, output_filepath='covid19_ai_imaging.jsonl')

You can also use dump_queries to iterate over a bunch of queries for all available databases.

from paperscraper import dump_queries

queries = [[covid19, ai, mi], [covid19, ai], [ai]]
dump_queries(queries, '.')
  • Scrape papers from Google Scholar:

Thanks to scholarly, there is an endpoint for Google Scholar too. It does not understand Boolean expressions like the others, but should be used just like the Google Scholar search fields.

from paperscraper.scholar import get_and_dump_scholar_papers
topic = 'Machine Learning'
get_and_dump_scholar_papers(topic)

Citation search

A plus of the Scholar endpoint is that the number of citations of a paper can be fetched:

from paperscraper.scholar import get_citations_from_title
title = 'Über formal unentscheidbare Sätze der Principia Mathematica und verwandter Systeme I.'
get_citations_from_title(title)

NOTE: The scholar endpoint does not require authentification but since it regularly prompts with captchas, it's difficult to apply large scale.

Journal impact factor

You can also retrieve the impact factor for all journals indexed by citefactor:

from paperscraper.journal_if import Impactor
i = Impactor()

Then, i.journal_to_if should give you a dictionary wit journal to IF mappings for >9000 journals as of 2014.

Plotting

When multiple query searches are performed, two types of plots can be generated automatically: Venn diagrams and bar plots.

Barplots

Compare the temporal evolution of different queries across different servers.

from paperscraper import QUERY_FN_DICT
from paperscraper.postprocessing import aggregate_paper
from paperscraper.utils import get_filename_from_query, load_jsonl

# Define search terms and their synonyms
ml = ['Deep learning', 'Neural Network', 'Machine learning']
mol = ['molecule', 'molecular', 'drug', 'ligand', 'compound']
gnn = ['gcn', 'gnn', 'graph neural', 'graph convolutional', 'molecular graph']
smiles = ['SMILES', 'Simplified molecular']
fp = ['fingerprint', 'molecular fingerprint', 'fingerprints']

# Define queries
queries = [[ml, mol, smiles], [ml, mol, fp], [ml, mol, gnn]]

root = '../keyword_dumps'

data_dict = dict()
for query in queries:
    filename = get_filename_from_query(query)
    data_dict[filename] = dict()
    for db,_ in QUERY_FN_DICT.items():
        # Assuming the keyword search has been performed already
        data = load_jsonl(os.path.join(root, db, filename))

        # Unstructured matches are aggregated into 6 bins, 1 per year
        # from 2015 to 2020. Sanity check is performed by having 
        # `filtering=True`, removing papers that don't contain all of
        # the keywords in query.
        data_dict[filename][db], filtered = aggregate_paper(
            data, 2015, bins_per_year=1, filtering=True,
            filter_keys=query, return_filtered=True
        )

# Plotting is now very simple
from paperscraper.plotting import plot_comparison

data_keys = [
    'deeplearning_molecule_fingerprint.jsonl',
    'deeplearning_molecule_smiles.jsonl', 
    'deeplearning_molecule_gcn.jsonl'
]
plot_comparison(
    data_dict,
    data_keys,
    title_text="'Deep Learning' AND 'Molecule' AND X",
    keyword_text=['Fingerprint', 'SMILES', 'Graph'],
    figpath='mol_representation'
)

molreps

Venn Diagrams

from paperscraper.plotting import (
    plot_venn_two, plot_venn_three, plot_multiple_venn
)

sizes_2020 = (30842, 14474, 2292, 35476, 1904, 1408, 376)
sizes_2019 = (55402, 11899, 2563)
labels_2020 = ('Medical\nImaging', 'Artificial\nIntelligence', 'COVID-19')
labels_2019 = ['Medical Imaging', 'Artificial\nIntelligence']

plot_venn_two(sizes_2019, labels_2019, title='2019', figname='ai_imaging')

2019

plot_venn_three(
    sizes_2020, labels_2020, title='2020', figname='ai_imaging_covid'
)

2020)

Or plot both together:

plot_multiple_venn(
    [sizes_2019, sizes_2020], [labels_2019, labels_2020], 
    titles=['2019', '2020'], suptitle='Keyword search comparison', 
    gridspec_kw={'width_ratios': [1, 2]}, figsize=(10, 6),
    figname='both'
)

both

Citation

If you use paperscraper, please cite the following:

@article{born2021on,
	title = {On the role of artificial intelligence in medical imaging of COVID-19},
	journal = {Patterns},
	volume = {2},
	number = {6},
	pages = {100269},
	year = {2021},
	issn = {2666-3899},
	url = {https://doi.org/10.1016/j.patter.2021.100269},
	author = {Jannis Born and David Beymer and Deepta Rajan and Adam Coy and Vandana V. Mukherjee and Matteo Manica and Prasanth Prasanna and Deddeh Ballah and Michal Guindy and Dorith Shaham and Pallav L. Shah and Emmanouil Karteris and Jan L. Robertus and Maria Gabrani and Michal Rosen-Zvi}
}

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

paperscraper-0.2.2.tar.gz (26.6 kB view hashes)

Uploaded Source

Built Distribution

paperscraper-0.2.2-py3-none-any.whl (31.0 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page