Train classifiers with Whatsapp chat data and predict the author of new messages
whosapp does not need to be installed, but has the following dependencies:
- Python > 2.7
- Python package
scikit-learn - Python package
pandas
whosapp is a Python script that can either train and save a model, or make predictions with a previously trained model. Train a model with the option --train:
python whosapp.py --train --model model.pkl --data training.txt
If --data is omitted, the trainer assumes training data is supplied from STDIN. The option --model is required in any case.
After training a model, use it for predictions with --predict:
echo "completely new message! You'll never guess who wrote me." | python whosapp.py --predict --model model.pkl
Alternatively, you can specify a file with a list of samples for which a class should be predicted with the option --samples.
Type python whosapp.py --help for more details and advanced options.
The program works directly on the data dump currently provided by Whatsapp. On your phone, find a chat with at least two members, and have the contents emailed to you as a *.txt file:
https://faq.whatsapp.com/fil/android/23756533
This file can be used directly with the --data option for training. The chat should contain a sizeable number of messages, do not try anything below 1000 messages.
The names predicted by a whosapp model will correspond to the names people have on your phone. If that's inconvenient, use the option --rename-authors.