Usually when extracting text from open source OCRs like Tesseract, we're most likely to encounter linked words duo to OCR quality extraction.
For example : instead of extracting "Très bon service", one might get sudenlly "Très bonservice". So when doing feature engineering with BOW, TFIDF or even word2vec models, the algorithm will consider that "bonservice" as a unique feature, while it is not.
To deal with this problem, I built a module dealing with semantic word segmentation without any predefined corpus.
Use the package manager pip to install fr_word_segment.
pip3 install fr-word-segment
python3 -m spacy download frfrom fr_word_segment import wordseg
# suppose that a french spellchecker detect this token as misspelled
token = "soitmoinscompliqué"
# apply segmentation function on the given token
result = wordseg.segment_token(token)
# show results
print("raw token is {}".format(token)) # "soitmoinscompliqué"
print("processed token is {}".format(result)) # "soit moins compliqué"Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.