clean-text 0.7.1

Functions to preprocess and normalize text.

clean-text

User-generated content on the Web and in social media is often dirty. Preprocess your scraped data with clean-text to create a normalized text representation. For instance, turn this corrupted input:

A bunch of \\u2018new\\u2019 references, including [Moana](https://en.wikipedia.org/wiki/Moana_%282016_film%29).


»Yóù àré     rïght <3!«

into this clean output:

A bunch of 'new' references, including [moana](<URL>).

"you are right <3!"

clean-text uses ftfy, unidecode and numerous hand-crafted rules, i.e., RegEx.

Installation

To install the GPL-licensed package unidecode alongside:

pip install clean-text[gpl]

You may want to abstain from GPL:

pip install clean-text

NB: This package is named clean-text and not cleantext.

If unidecode is not available, clean-text will resort to Python's unicodedata.normalize for transliteration. Transliteration to closest ASCII symbols involes manually mappings, i.e., ê to e. unidecode's mapping is superiour but unicodedata's are sufficent. However, you may want to disable this feature altogether depending on your data and use case.

To make it clear: There are inconsistencies between processing text with or without unidecode.

Usage

from cleantext import clean

clean("some input",
    fix_unicode=True,               # fix various unicode errors
    to_ascii=True,                  # transliterate to closest ASCII representation
    lower=True,                     # lowercase text
    no_line_breaks=False,           # fully strip line breaks as opposed to only normalizing them
    no_code=False,                  # replace all code snippets with a special token
    no_urls=False,                  # replace all URLs with a special token
    no_emails=False,                # replace all email addresses with a special token
    no_phone_numbers=False,         # replace all phone numbers with a special token
    no_ip_addresses=False,          # replace all IP addresses with a special token
    no_file_paths=False,            # replace all file paths with a special token
    no_numbers=False,               # replace all numbers with a special token
    no_digits=False,                # replace all digits with a special token
    no_currency_symbols=False,      # replace all currency symbols with a special token
    no_punct=False,                 # remove punctuations
    replace_with_punct="",          # instead of removing punctuations you may replace them
    exceptions=None,                # list of regex patterns to preserve verbatim
    replace_with_code="<CODE>",
    replace_with_url="<URL>",
    replace_with_email="<EMAIL>",
    replace_with_phone_number="<PHONE>",
    replace_with_ip_address="<IP>",
    replace_with_file_path="<FILE_PATH>",
    replace_with_number="<NUMBER>",
    replace_with_digit="0",
    replace_with_currency_symbol="<CUR>",
    lang="en"                       # set to 'de' for German special handling
)

Carefully choose the arguments that fit your task. The default parameters are listed above.

Preserving patterns with exceptions

Use exceptions to protect specific text patterns from being modified during cleaning. Each entry is a regex pattern string; all matches are preserved verbatim (not lowered, not transliterated — exactly as they appeared in the input).

from cleantext import clean

# Preserve a literal compound word while removing other punctuation
clean("drive-thru and text---cleaning", no_punct=True, exceptions=["drive-thru"])
# => 'drive-thru and textcleaning'

# Preserve all hyphenated compound words using a regex
clean("drive-thru and pick-up", no_punct=True, exceptions=[r"\w+-\w+"])
# => 'drive-thru and pick-up'

# Multiple exception patterns
clean("drive-thru costs $5", no_punct=True, no_currency_symbols=True,
      exceptions=[r"\w+-\w+", r"\$\d+"])
# => 'drive-thru costs $5'

You may also only use specific functions for cleaning. For this, take a look at the source code.

Cleaning multiple texts in parallel

Use clean_texts() to clean a list of strings. Set n_jobs to enable parallel processing via Python's built-in multiprocessing:

from cleantext import clean_texts

# Sequential (default) — no multiprocessing overhead
clean_texts(["text one", "text two", "text three"])

# Use all available CPU cores
clean_texts(["text one", "text two", "text three"], n_jobs=-1)

# Use a specific number of workers
clean_texts(["text one", "text two", "text three"], n_jobs=4)

# All clean() keyword arguments are supported
clean_texts(texts, n_jobs=-1, no_urls=True, lang="de", lower=False)

n_jobs semantics:

• 1 or None — sequential processing (default, zero overhead)
• -1 — use all available CPU cores
• -2 — use all cores except one, etc.
• Any positive integer — use exactly that many workers
• 0 — raises ValueError

Supported languages

So far, only English and German are fully supported. It should work for the majority of western languages. If you need some special handling for your language, feel free to contribute. 🙃

Using clean-text with scikit-learn

There is also scikit-learn compatible API to use in your pipelines. All of the parameters above work here as well.

pip install clean-text[gpl,sklearn]
pip install clean-text[sklearn]

from cleantext.sklearn import CleanTransformer

cleaner = CleanTransformer(no_punct=False, lower=False)

cleaner.transform(['Happily clean your text!', 'Another Input'])

Development

Use poetry.

See RELEASING.md for how to publish a new version.

Contributing

If you have a question, found a bug or want to propose a new feature, have a look at the issues page.

Pull requests are especially welcomed when they fix bugs or improve the code quality.

If you don't like the output of clean-text, consider adding a test with your specific input and desired output.

Related Work

Generic text cleaning packages

• https://github.com/pudo/normality
• https://github.com/davidmogar/cucco
• https://github.com/lyeoni/prenlp
• https://github.com/s/preprocessor
• https://github.com/artefactory/NLPretext
• https://github.com/cbaziotis/ekphrasis

Full-blown NLP libraries with some text cleaning

• https://github.com/chartbeat-labs/textacy
• https://github.com/jbesomi/texthero

Remove or replace strings

• https://github.com/vi3k6i5/flashtext
• https://github.com/ddelange/retrie

Detect dates

• https://github.com/scrapinghub/dateparser

Clean massive Common Crawl data

• https://github.com/facebookresearch/cc_net

Acknowledgements

Built upon the work by Burton DeWilde for Textacy.

License

Apache