PHASE: Learning Emotional Phase-aware Representations for Suicide Ideation Detection on Social Media

Abstract

Recent psychological studies indicate that individuals exhibiting suicidal ideation increasingly turn to social media rather than mental health practitioners. Contextualizing the buildup of such ideation is critical for the identification of users at risk. In this work, we focus on identifying suicidal intent in tweets by augmenting linguistic models with emotional phases modeled from users' historical context. We propose PHASE, a time-and phase-aware framework that adaptively learns features from a user's historical emotional spectrum on Twitter for preliminary screening of suicidal risk. Building on clinical studies, PHASE learns phase-like progressions in users' historical Plutchik-wheel-based emotions to contextualize suicidal intent. While outperforming stateof-the-art methods, we show the utility of temporal and phase-based emotional contextual cues for suicide ideation detection. We further discuss practical and ethical considerations. 1

Capturing Changes in Mood Over Time in Longitudinal Data Using Ensemble Methodologies

“…Furthermore, people who attempt suicide have a higher proportion of emotional posts, increasing after the incident. In line with these findings, several works are modelling the emotional information found in the online discourse of users for classifying the suicide risk (Ji et al, 2021;Sawhney et al, 2021;Bitew et al, 2019;.…”

“…Using these markers, the authors can predict which individuals are more prone to express suicide ideation in future posts. Through a time-aware approach, Sawhney et al (2021) propose a framework that uses people's historical and emotional spectrum when assessing the risk of a specific post. Tsakalidis et al (2022b) propose to take the temporal information into account by identifying the changes in people's behaviour and mood on social media.…”

Identifying Moments of Change from Longitudinal User Text

“…For example, Cao et al (2019) encode microblog posts using suicide-oriented embeddings fed to an LSTM network to assess the suicidality risk at post level. Sawhney et al (2020bSawhney et al ( , 2021 improves further on predicting suicidality at postlevel by jointly considering an emotion-oriented post representation and the user's emotional state as reflected through their posting history with temporally aware models. The recent shared tasks in eRisk also consider sequences of user posts in order to classify a user as a "positive" (wrt self-harm or pathological gambling) or "control" case (Losada et al, 2020;Parapar et al, 2021).…”

WWBP-SQT-lite: Multi-level Models and Difference Embeddings for Moments of Change Identification in Mental Health Forums

“…Beyond utilizing the best transformer based approaches, we also explore relevant theoretical features to understand the relationship between moment of change and psychological/demographic constructs. Furthermore, recent works (Sawhney et al, 2020(Sawhney et al, , 2021 have shown the importance of joint modelling of such theoretical dimensions with the present-day neural approaches.…”