Skip to content

Latest commit

 

History

History
147 lines (103 loc) · 5.14 KB

README.md

File metadata and controls

147 lines (103 loc) · 5.14 KB
Raw

CleanText

CleanText is a Python package that I use in my research projects to clean social media captions but will likely be useful to others beyond that scope, so I wanted to make it available here as well.

Dependencies

The package is dependent on some various other packages: html2text, yaml, and unidecode. They need all to be installed:

pip install html2text
pip install yaml
pip install unidecode

Usage

To import the package, make sure that you run the following command:

from CleanText import *

Next, set up an object with the default settings (see below for instructions on how to change the settings):

text = "This is an advertisement to call # 222-109-1100 or go to http://www.apple.com. Maybe this will be replaced with definitely and perhaps also. If you're using quotation marks, those will be replaced."
cleaner = CleanText(text)

To access the cleaned text, simply call:

cleaner.text

If you surround it with a print() function, it should generate the following result:

advertisement call go maybe replaced definitely perhaps also

Ingestation

A CleanText instance accepts three arguments. text, as seen above, is a required argument. You can also provide your own list of stopwords as well as replacements.

Stopwords

A list of strings provided as stopwords will expand the numbers of words that will be removed in the stopword removal process. For example:

text = "This is an advertisement to call # 222-109-1100 or go to http://www.apple.com. Maybe this will be replaced with definitely and perhaps also. If you're using quotation marks, those will be replaced."
cleaner = CleanText(text, stopwords=['replaced', 'advertisement', 'call'])

Once again calling print(cleaner.text) will provide you with a different result:

go maybe definitely perhaps also using quotation marks

Custom replacements

If you want to make sure that certain words get changed, perhaps from one verb form to another or something similar, you can do so using the special_replacements dictionary that can be provided in the ingestion phase. For example, using the same text variable as before:

cleaner = CleanText(text, special_replacements={'advertisement': 'ad', 'quotation': 'bracket', 'call': 'give a ring'})

Once again calling print(cleaner.text) will provide you with a different result:

ad give ring go maybe replaced definitely perhaps also using bracket marks replaced

Note how the a from the "give a ring" in the replacement from the special_replacement dictionary has been removed. This is because the stopwords algorithm runs after your replacements have been made.

General settings

The standard settings in the package will:

  • modify to your text:
    • make your text lowercase
    • expand contractions
    • remove stopwords
  • remove elements of your text:
    • any hyperlinks
    • any digits
    • any emojis
    • any hashtags
    • any mentions (@usernames)
    • any punctuation

Any of those standards can be changed by modifying the "Standard Settings" at the top of the module file (CleanText.py):

Setting If set to True
LOWER Modifies text to lowercase.
EXPAND_CONTRACTIONS Modifies text to expand any contractions (ex. can't, won't).
REMOVE_STOPWORDS Modifies text to remove commonly occurring stopwords (ex. I, you, me).
LINKS Removes any links in the text.
DIGITS Removes any digits in the text.
EMOJI Removes any emojies from the text.
HASH Removes any hashtags from the text.
AT Removes any mentions of users from the text (ex. @kallewesterling)
PUNCTUATION Removes any punctuation from the text.

You can also change those settings for each instantiation. Building on the example from above:

cleaner = CleanText(text)
cleaner.lower = False
cleaner.clean()

Note that if you change any settings this way, you need to re-run the CleanText.clean() method again, to clean the text that was ingested as the instance was created.

To put what we learn above to the test, we can run:

text = 'This is an Advertisement to Call # 222-109-1100 or go to http://www.apple.com. Maybe this will be replaced with definitely and perhaps also.'
cleaner = CleanText(text)
first_round = cleaner.text

cleaner.lower = False
cleaner.clean()
second_round = cleaner.text

print("Before:", first_round)
print("After:", second_round)

This should generate the following results:

Before: advertisement call go maybe replaced definitely perhaps also
After: This Advertisement Call go Maybe replaced definitely perhaps also

Future improvements

In the future, I want to remove the dependency on NLTK for stopwords and use a different, smaller package for this process. NLTK is too heavy for any users who do not need it. The current version tries a different approach, borrowed from YoastSEO.