Automated detection of substance use-related social media posts based on hybrid GAT-BiGRU model
R. Nadanasabai, Venkata Ramaiah Turlapati, Ramesh Kumar Miryala, P. L. Swerna, M. Nimisha, R. Arun · 2025
It is possible that social media can shed light on substance addiction and abuse. This study set out to apply a deep learning algorithm we developed to automatically classify Instagram users’ risk for alcohol, tobacco, and drug usage. A total of three thousand active Instagram users were surveyed. Automated estimation techniques can be useful for future generations of population-level risk assessment and intervention delivery. Preprocessing, learning features, and training the model are the three steps that make up the proposed technique. While LDA and M-DW are utilized for feature learning, TF-IDF is employed for preprocessing. When training the model, various methods were utilized, including BiGRU, DNN-BiGRU, and GAT-BiGRU. When compared to two more traditional methods, GAT-BiGRU triumphs.