Identifying Moments of Change from Longitudinal User Text

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 4647 - 4660 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Identifying Moments of Change from Longitudinal User Text Adam Tsakalidis1,2, Federico Nanni2, Anthony Hills1, Jenny Chim1, Jiayu Song1, Maria Liakata1,2,3 1 Queen Mary University of London, London, United Kingdom 2 The Alan Turing Institute, London, United Kingdom 3 University of Warwick, Coventry, United Kingdom a.tsakalidis;m.liakata@qmul.ac.uk Abstract Identifying changes in individuals’ behaviour and mood, as observed via content shared on online platforms, is increasingly gaining im- portance. Most research to-date on this topic focuses on either: (a) identifying individuals at risk or with a certain mental health condition given a batch of posts or (b) providing equiva- lent labels at the post level. A disadvantage of such work is the lack of a strong temporal com- ponent and the inability to make longitudinal assessments following an individual’s trajec- tory and allowing timely interventions. Here we define a new task, that of identifying mo- ments of change in individuals on the basis of their shared content online. The changes we consider are sudden shifts in mood (switches) or gradual mood progression (escalations). We have created detailed guidelines for capturing moments of change and a corpus of 500 man- ually annotated user timelines (18.7K posts). We have developed a variety of baseline models drawing inspiration from related tasks and show that the best performance is obtained through context aware sequential modelling. We also introduce new metrics for capturing rare events in temporal windows. 1 Introduction Linguistic and other content from social media data has been used in a number of different studies to obtain biomarkers for mental health. This is gain- ing importance given the global increase in men- tal health disorders, the limited access to support services and the prioritisation of mental health as an area by the World Health Organization (2019). Studies using linguistic data for mental health focus on recognising specific conditions related to men- tal health (e.g., depression, bipolar disorder) (Hus- seini Orabi et al., 2018), or identifying self-harm ideation in user posts (Yates et al., 2017; Zirikly et al., 2019). However, none of these works, even when incorporating a notion of time (Lynn et al., Figure 1: Example of an Escalation (with a darker “peak”) and a Switch within a user’s timeline. 2018; Losada et al., 2020), identify how an individ- ual’s mental health changes over time. Yet being able to make assessments on a longitudinal level from linguistic and other digital content is impor- tant for clinical outcomes, and especially in mental health (Velupillai et al., 2018). The ability to detect changes in individual’s mental health over time is also important in enabling platform moderators to prioritise interventions for vulnerable individuals (Wadden et al., 2021). Users who currently engage with platforms and apps for mental health support (Neary and Schueller, 2018) would also benefit from being able to monitor their well-being in a longitudinal manner. Motivated by the lack …

Content truncated to 500 words