Developed by Fast Data Science, https://fastdatascience.com
Source code at https://github.com/fastdatascience/company_named_entity_recognition
Tutorial at https://fastdatascience.com/company-named-entity-recognition-python-library/
This is a lightweight Python library for finding company names in a string, otherwise known as named entity recognition (NER) and named entity linking.
Please note this library finds only high confidence companies.
It also only finds the English names of these companies. Names in other languages are not supported.
It also doesn't find short code names of companies, such as abbreviations commonly used in medicine, such as "Ceph" for "Cephradin" - as these are highly ambiguous.
Interested in other kinds of named entity recognition (NER)? Finances, company names, countries, locations, proteins, genes, molecules?
If your NER problem is common across industries and likely to have been seen before, there may be an off-the-shelf NER tool for your purposes, such as our Country Named Entity Recognition Python library. Dictionary-based named entity recognition is not always the solution, as sometimes the total set of entities is an open set and can't be listed (e.g. personal names), so sometimes a bespoke trained NER model is the answer. For tasks like finding email addresses or phone numbers, regular expressions (simple rules) are sufficient for the job.
If your named entity recognition or named entity linking problem is very niche and unusual, and a product exists for that problem, that product is likely to only solve your problem 80% of the way, and you will have more work trying to fix the final mile than if you had done the whole thing manually. Please contact Fast Data Science and we'll be glad to discuss. For example, we've worked on a consultancy engagement to find molecule names in papers, and match author names to customers where the goal was to trace molecule samples ordered from a pharma company and identify when the samples resulted in a publication. For this case, there was no off-the-shelf library that we could use.
For a problem like identifying country names in English, which is a closed set with well-known variants and aliases, and an off-the-shelf library is usually available.
For identifying a set of molecules manufactured by a particular company, this is the kind of task more suited to a consulting engagement.
Python 3.9 and above
You can contact Thomas Wood or Fast Data Science team at https://fastdatascience.com/.
You can install from PyPI.
pip install company-named-entity-recognition
You must first tokenise your input text using a tokeniser of your choice (NLTK, spaCy, etc).
You pass a list of strings to the find_companies
function.
Example 1
from company_named_entity_recognition import find_companies
find_companies("i bought some Prednisone".split(" "))
outputs a list of tuples.
[({'name': 'Prednisone', 'synonyms': {'Sone', 'Sterapred', 'Deltasone', 'Panafcort', 'Prednidib', 'Cortan', 'Rectodelt', 'Prednisone', 'Cutason', 'Meticorten', 'Panasol', 'Enkortolon', 'Ultracorten', 'Decortin', 'Orasone', 'Winpred', 'Dehydrocortisone', 'Dacortin', 'Cortancyl', 'Encorton', 'Encortone', 'Decortisyl', 'Kortancyl', 'Pronisone', 'Prednisona', 'Predniment', 'Prednisonum', 'Rayos'}, 'medline_plus_id': 'a601102', 'mesh_id': 'D018931', 'companybank_id': 'DB00635'}, 3, 3)]
You can ignore case with:
find_companies("i bought some prednisone".split(" "), is_ignore_case=True)
The Company Named Entity Recognition library is independent of other NLP tools and has no dependencies. You don't need any advanced system requirements and the tool is lightweight. However, it combines well with other libraries such as spaCy or the Natural Language Toolkit (NLTK).
Here is an example call to the tool with a spaCy Doc object:
from company_named_entity_recognition import find_companies
import spacy
nlp = spacy.blank("en")
doc = nlp("i routinely rx rimonabant and pts prefer it")
find_companies([t.text for t in doc], is_ignore_case=True)
outputs:
[({'name': 'Rimonabant', 'synonyms': {'Acomplia', 'Rimonabant', 'Zimulti'}, 'mesh_id': 'D063387', 'companybank_id': 'DB06155'}, 3, 3)]
You can also use the tool together with the Natural Language Toolkit (NLTK):
from company_named_entity_recognition import find_companies
from nltk.tokenize import wordpunct_tokenize
tokens = wordpunct_tokenize("i routinely rx rimonabant and pts prefer it")
find_companies(tokens, is_ignore_case=True)
The main data source is from Companybank, augmented by datasets from the NHS, MeSH, Medline Plus and Wikipedia.
If you want to update the dictionary, you can use the data dump from Companybank and replace the file companybank vocabulary.csv
:
- Download the open data dump from https://go.companybank.com/releases/latest#open-data
If you want to update the Wikipedia dictionary, download the dump from:
and run extract_company_names_and_synonyms_from_wikipedia_dump.py
If you want to update the dictionary, run
python download_mesh_dump_and_extract_company_names_and_synonyms.py
This will download the latest XML file from NIH.
If the link doesn't work, download the open data dump manually from https://www.nlm.nih.gov/. It should be called something like desc2023.xml
. And comment out the Wget/Curl commands in the code.
- Data from Companybank is licensed under CC0.
To the extent possible under law, the person who associated CC0 with the CompanyBank Open Data has waived all copyright and related or neighboring rights to the CompanyBank Open Data. This work is published from: Canada.
- Text from Wikipedia data dump is licensed under GNU Free Documentation License and Creative Commons Attribution-Share-Alike 3.0 License. More information.
If you'd like to contribute to this project, you can contact us at https://fastdatascience.com/ or make a pull request on our Github repository. You can also raise an issue.
Test code is in tests/ folder using unittest.
The testing tool tox
is used in the automation with GitHub Actions CI/CD.
Install tox and run it:
pip install tox
tox
In our configuration, tox runs a check of source distribution using check-manifest (which requires your repo to be git-initialized (git init
) and added (git add .
) at least), setuptools's check, and unit tests using pytest. You don't need to install check-manifest and pytest though, tox will install them in a separate environment.
The automated tests are run against several Python versions, but on your machine, you might be using only one version of Python, if that is Python 3.9, then run:
tox -e py39
Thanks to GitHub Actions' automated process, you don't need to generate distribution files locally. But if you insist, click to read the "Generate distribution files" section.
This package is based on the template https://pypi.org/project/example-pypi-package/
This package
- uses GitHub Actions for both testing and publishing
- is tested when pushing
master
ormain
branch, and is published when create a release - includes test files in the source distribution
- uses setup.cfg for version single-sourcing (setuptools 46.4.0+)
The code to re-release Harmony on PyPI is as follows:
source activate py311
pip install twine
rm -rf dist
python setup.py sdist
twine upload dist/*
The tool was developed:
- Thomas Wood (Fast Data Science)
MIT License. Copyright (c) 2023 Fast Data Science
Wood, T.A., Company Named Entity Recognition [Computer software], Version 1.0.1, accessed at https://fastdatascience.com/company-named-entity-recognition-python-library, Fast Data Science Ltd (2023)
@unpublished{companynamedentityrecognition,
AUTHOR = {Wood, T.A.},
TITLE = {Company Named Entity Recognition (Computer software), Version 1.0.1},
YEAR = {2023},
Note = {To appear},
}