Proof-of-concept to determine potential suicide/depression risk from text. Idea was developed by a collaboration between myself and Ebrahim Payberah
This repository has a Dataset to load into google collab. 20,000 suicide/non-suicide train (even split), 10,000 suicide/non-suicide test (even split).
Original dataset source on kaggle by Nikhileswar Komati. Cleaned and formatted for pytorch
Original GPT-2 text classification by George Mihaila
sd_parser.py was used to generate sd.tar.xz from Suicide_Detection.csv (not provided) from the kaggle link above.
Collab notebook located here
While training model collab does have prediction at the end, if you wish to just load pre-trained model and predict from that, open this collab notebook instead