This repo holds the code for the modeling from the paper Opioid Death Projections with AI-based Forecasts using Social Media Language, which proposes the Transformer for Opioid Prediction model.
Data used for training TrOP is included as a CSV file in the data directory. It contains the 7 socio-economic status variables, yearly opioid mortality, and the dimensionality reduced topic vectors (20 dimensions). If you are looking for the raw (non-reduced) yearly topic vectors please look for them in the County Tweet Lexical Bank.
@article{matero_opioid_2023,
title = {Opioid death projections with {AI}-based forecasts using social media language},
volume = {6},
issn = {2398-6352},
url = {https://doi.org/10.1038/s41746-023-00776-0},
doi = {10.1038/s41746-023-00776-0},
number = {1},
journal = {npj Digital Medicine},
author = {Matero, Matthew and Giorgi, Salvatore and Curtis, Brenda and Ungar, Lyle H. and Schwartz, H. Andrew},
month = mar,
year = {2023},
pages = {35},
}