Emotionally-Informed Models for Detecting Moments of Change and Suicide Risk Levels in Longitudinal Social Media Data

Abstract

In this shared task, we focus on detecting mental health signals in Reddit users' posts through two main challenges: A) capturing mood changes (anomalies) from the longitudinal set of posts (called timelines), and B) assessing the users' suicide risk-levels. Our approaches leverage emotion recognition on linguistic content by computing emotion/sentiment scores using pre-trained BERTs on users' posts and feeding them to machine learning models, including XGBoost, Bi-LSTM, and logistic regression. For Task-A, we detect longitudinal anomalies using a sequence-to-sequence (seq2seq) autoencoder and capture regions of mood deviations. For Task-B, our two models utilize the BERT emotion/sentiment scores. The first computes emotion bandwidths and merges them with n-gram features, and employs logistic regression to detect users' suicide risk levels. The second model predicts suicide risk on the timeline level using a Bi-LSTM on Task-A results and sentiment scores. Our results outperformed most participating teams and ranked in the top three in Task-A. In Task-B, our methods surpass all others and return the best macro and micro F1 scores.

Overview of the CLPsych 2022 Shared Task: Capturing Moments of Change in Longitudinal User Posts

“…WResearch (Bayram and Benhiba, 2022) completed 4/5 submissions in the DE. In Task A, they derived emotionally-informed vectors from pretrained models and constructed abnormality vectors (i.e., differences in expected vs predicted vectors via a seq2seq model) and differences in the vectors of consecutive posts, using them as inputs to post-level classifiers that take into account the class imbalance.…”

Proceedings of the Eighth Workshop on Computational Linguistics and Clinical Psychology

“…WResearch (Bayram and Benhiba, 2022) completed 4/5 submissions in the DE. In Task A, they derived emotionally-informed vectors from pretrained models and constructed abnormality vectors (i.e., differences in expected vs predicted vectors via a seq2seq model) and differences in the vectors of consecutive posts, using them as inputs to post-level classifiers that take into account the class imbalance.…”