This repository stores the deep learning based code for Suicide Risk Assessment with Multi-level Dual-Context Language and BERT, SBU-HLAB's 2019 CLPsych submission. All non-deep learning models used the DLATK package for quick iteration on logistic regression models.
If you have any questions regarding the paper please contact mmatero -at- cs [dot] stonybrook . [edu] (PhD student) or has -at- cs [dot] stonybrook . [edu] (Lab director).
The repo holds the PyTorch model defintiion of our attenion-based LSTM network, which scored on a F1 of .50 on Task A data. Alternatively, one could use this class file to instantiate a dual-context variant, used for task B, which uses Task A data (suicide-context) and task C data (non-suicide context) as described in the paper.
@inproceedings{matero2019suicide,
title={Suicide risk assessment with multi-level dual-context language and BERT},
author={Matero, Matthew and Idnani, Akash and Son, Youngseo and Giorgi, Salvatore and Vu, Huy and Zamani, Mohammad and Limbachiya, Parth and Guntuku, Sharath Chandra and Schwartz, H Andrew},
booktitle={Proceedings of the Sixth Workshop on Computational Linguistics and Clinical Psychology},
pages={39--44},
year={2019}
}