Integrated Multi factor Trajectory Deep Learning Model for Suicide Ideation Detection from Social Media Interactions

Abstract

Depression and suicidal ideation (SI) have become very common among youths recently. Traditional methods of detecting SI through interaction with medical practitioners is not effective for earlier diagnosis. Recently detecting SI from social media interactions has gained prominence. Many deep learning techniques have been proposed to predict SI from texts. Most of the current methods analyze the texts for depression cues and their emotional trajectory to predict SI. They don’t use the knowledge about various inherent multi criteria factors like personality trait, degree of hopelessness, trait impulsivity, abuse, conflicts, reasons for living spread across messages and their temporal trajectories in SI prediction. Integrating this multi criteria factors based temporal trajectories in SI prediction will increase the prediction accuracy and reduce false positives. This work addresses this gap and proposes an integrated multi factor trajectory deep learning model for SI prediction from social media interactions. The solution extracts multiple risk and protective factor-based features from social media texts. The features are organized to trajectory vectors and classified by parallel cascaded LSTM to SI risk. The solution proposed has at least 2% higher accuracy in comparison to the most recent LSTM attention-based works when tested against Twitter dataset.