parisfellows_anonyme

Automatically pseudo-anonymise name of people in Cour des Comptes's jurisprudence

• We explore 138 documents.
• We have more than 12 k different words.
• We have more 420 k words (with 3147 positive / others are negative)

How to :

Donwload data from this link then dezip it. You should see a directory dataon root.

**Run script : **

• python reading_doc_files.py --> Create data.csv file with all features and structure
• python trainning.py --> Train the model and give some metrics
• get_prediction.py --> Read & processs a .docx (line 220) to anonymise it in ouput directory.

Create ouput files :

• [name_of_file]_log.csv : Log of this file (warning is a bool)
• [name_of_file].txt : Return the text with anonymise result.
• [name_of_file].html : Return the text in html balise with color (green seems OK, Red mean warning this could be a error).

result of html file :