text-scrubber is a Python package that offers text scrubbing functionality, providing building blocks for string cleaning as well as normalizing geographical text (countries/states/cities).
Full documentation is available at https://sybrenjansen.github.io/text-scrubber/.
The TextScrubber class cleans a single or a collection of strings. It can be easily constructed and configured with building blocks:
from text_scrubber import TextScrubber
ts = (TextScrubber().to_ascii()
.lowercase()
.tokenize()
.remove_stop_words()
.join())which can then be used as:
ts.transform('héLlô there, WòrlD') # outputs 'hello world'or with an iterable of input:
ts.transform(['héLlô there, WòrlD', 'slímm̀er ÀI']) # outputs ['hello world', 'slimmer AI']For a complete list of building blocks please refer to the TextScrubber API reference.
The text_scrubber.geo module contains functions to normalize geographical data which deal with spelling errors, country name variations, etc.:
from text_scrubber.geo import normalize_country, normalize_region, normalize_city
"""
Countries
"""
normalize_country('Peoples rep. of China')
# [Location(canonical_name='China', matched_name='Peoples Republic of China', country=None,
# score=1.0)]
normalize_country('Deutschland')
# [Location(canonical_name='Germany', matched_name='Deutschland', country=None, score=1.0)]
normalize_country('st Nevis and Kitties')
# [Location(canonical_name='Saint Kitts and Nevis', matched_name='Saint Kitts and Nevis',
# country=None, score=0.75)]
normalize_country('ira')
# [Location(canonical_name='Iran', matched_name='Iran', country=None, score=0.857...),
# Location(canonical_name='Iraq', matched_name='Iraq', country=None, score=0.857...)]
"""
Cities
"""
normalize_city('Leibnitz', ['Austria'])
# [Location(canonical_name='Leibnitz', matched_name='Leibnitz', country='Austria', score=1.0)]
normalize_city('heidelberg')
# [Location(canonical_name='Heidelberg', matched_name='Heidelberg', country='Germany',
# score=1.0),
# Location(canonical_name='Heidelberg', matched_name='Heidelberg', country='South Africa',
# score=1.0),
# Location(canonical_name='Heidelberg', matched_name='Heidelberg', country='United States',
# score=1.0)]
normalize_city('ohioo', ['US'])
# [Location(canonical_name='Ohio', matched_name='Ohio', country='United States',
# score=0.888...)]
normalize_city('Madri', ['Spain', 'US', 'Brazil'])
# [Location(canonical_name='Madrid', matched_name='Madrid', country='Spain',
# score=0.909...),
# Location(canonical_name='Madrid', matched_name='Madrid', country='United States',
# score=0.909...),
# Location(canonical_name='Mari', matched_name='Mari', country='Brazil',
# score=0.888...)]
"""
Regions
"""
normalize_region('triangle park', ['US'])
# [Location(canonical_name='The Triangle Park', matched_name='The Triangle Park',
# country='United States', score=1.0)]
normalize_region('Fur', ['Denmark'])
# [Location(canonical_name='Fur', matched_name='Fur', country='Denmark', score=1.0)]
normalize_region('texel', ['NL'])
# [Location(canonical_name='Texel', matched_name='Texel', country='Netherlands', score=1.0)]Each of the above normalization functions return the canonical name, matched name, the match score, and when normalizing cities or regions it will also contain the corresponding country. The difference between canonical and matched name stems from the fact that some countries, cities, or regions can have alternative names. E.g., NYC maps to New York City. When the query was NYCC the canonical name will be New York City, but the matched name NYC. The match scores are always between 0.0 and 1.0, where 1.0 is a perfect match. If a known mapping exists, like Deutschland to Germany, then the match score will be 1.0.
Note
When normalizing a country or finding countries in a string, the country attribute of a LocationMatch object is always None. The normalized name can be found using the canonical_name attribute.
The text_scrubber.geo module also contains functions to find the name of places (country, region, and city) in text dealing with spelling errors, country name variations, etc.:
from text_scrubber.geo import (find_city_in_string, find_country_in_string,
find_region_in_string)
"""
Countries
"""
find_country_in_string("Institute of German study, Accra, Ghana")
# [ExtractedLocation(location=Location(canonical_name='Ghana', matched_name='Ghana',
# country=None, score=1.0),
# substring='Ghana', substring_range=Range(start=34, end=39)),
# ExtractedLocation(location=Location(canonical_name='Germany', matched_name='Germany',
# country=None, score=0.923...),
# substring='German', substring_range=Range(start=13, end=19))]
find_country_in_string("Peking University, 5 Yiheyuan Rd, "
"Haidian District, Beijing, CH, 100871")
# This was a trick question though, as CH=Switzerland. China is CN
# [ExtractedLocation(location=Location(canonical_name='Switzerland', matched_name='CH',
# country=None, score=1.0),
# substring='CH', substring_range=Range(start=61, end=63))]
"""
Cities
"""
find_city_in_string("Météorage Pau France", {"France"})
# [ExtractedLocation(location=Location(canonical_name='Pau', matched_name='Pau',
# country='France', score=1.0),
# substring='Pau', substring_range=Range(start=10, end=13)),
# ExtractedLocation(location=Location(canonical_name='La Frasnée', matched_name='Фране',
# country='France', score=0.909...),
# substring='France', substring_range=Range(start=14, end=20))]
find_city_in_string("Bavarian Environment Agency, Hans Högn Straße 12, "
"95030 Hof Saale, Bavaria, Germany", {"Germany"})
# [ExtractedLocation(location=Location(canonical_name='Hof', matched_name='Hof',
# country='Germany', score=1.0),
# substring='Hof', substring_range=Range(start=56, end=59)),
# ExtractedLocation(location=Location(canonical_name='Saal', matched_name='Saal',
# country='Germany', score=0.888...),
# substring='Saale', substring_range=Range(start=60, end=65)),
# ExtractedLocation(location=Location(canonical_name='Trassem', matched_name='Trassem',
# country='Germany', score=0.857...),
# substring='Straße', substring_range=Range(start=39, end=45))]
"""
Regions
"""
find_region_in_string("Fur Museum, 7884 Fur, Denmark.", {"Denmark"})
# [ExtractedLocation(location=Location(canonical_name='Fur', matched_name='Fur',
# country='Denmark', score=1.0),
# substring='Fur', substring_range=Range(start=0, end=3)),
# ExtractedLocation(location=Location(canonical_name='Fur', matched_name='Fur',
# country='Denmark', score=1.0),
# substring='Fur', substring_range=Range(start=17, end=20)),
# ExtractedLocation(location=Location(canonical_name='Kingdom of Denmark',
# matched_name='Denmark', country='Denmark', score=1.0),
# substring='Denmark', substring_range=Range(start=22, end=29))]
find_region_in_string("Department of Biological Oceanography, Royal Netherlands Institute "
"for Sea Research (NIOZ), Texel, The Netherlands", {"Netherlands"})
# [ExtractedLocation(location=Location(canonical_name='Kingdom of the Netherlands',
# matched_name='Netherlands', country='Netherlands',
# score=1.0),
# substring='Netherlands', substring_range=Range(start=45, end=56)),
# ExtractedLocation(location=Location(canonical_name='Texel', matched_name='Texel',
# country='Netherlands', score=1.0),
# substring='Texel', substring_range=Range(start=92, end=97)),
# ExtractedLocation(location=Location(canonical_name='Kingdom of the Netherlands',
# matched_name='Netherlands', country='Netherlands',
# score=1.0),
# substring='Netherlands', substring_range=Range(start=103, end=114))]Note
Whenever a country is considered part of another country normalize_country will return the latter. E.g., Puerto Rico is mapped to United States and Greenland to Denmark.
Resources for cities and regions aren't all loaded when you import TextScrubber, they're loaded on the fly per country. This means that the first time you do a query it can take a while. The second time around the same query will be much faster, as will all other queries involving the same countr(y)(ies). You can load in resources per country in advance by using:
from text_scrubber.geo import (add_city_resources, add_region_resources,
normalize_country_to_country_codes)
country_codes = normalize_country_to_country_codes(['Netherlands', 'China', 'USA'])
add_city_resources(country_codes)
add_region_resources(country_codes, progress_bar=True)Note
Whenever a country is considered part of another country normalize_country_to_country_codes returns both.
There are clean functions available for countries/regions/cities, which all follow the same cleaning pipeline:
from text_scrubber.geo import clean_country, clean_region, clean_city
clean_country('cent afr rep.') # 'central african republic'
clean_region('Hyōgo') # 'hyogo'
clean_city('płońsk') # 'plonsk'
clean_city('neustadt/westerwald') # 'neustadt westerwald'If you want to build the documentation, please install the documentation dependencies by executing:
pip install .[docs]Documentation can be build by executing:
python setup.py build_docsDocumentation can also be build from the docs folder directly. In that case text-scrubber should be installed and available in your current working environment. Execute:
make htmlin the docs folder.