Capturing changes in mood over time
Increasingly the clinical community are looking for new and better diagnostic measures and mental health condition monitoring tools. Over the past decade, there has been a surge in methods at the intersection of NLP and mental health, showing that signals for the diagnosis of certain conditions can be found in language. However, most research tasks have been defined on the basis of classifying individuals (e.g., on the basis of suicide risk or on the basis of having a mental health condition or not), thus lacking the longitudinal component of monitoring an individual’s mood and well-being in real-time.
The CLPsych 2022 Shared Task
We introduce the problem of assessing changes in a person’s mood over time on the basis of their linguistic content. For the purpose of the task we focus on posting activity in online social media platforms. In particular, given a user’s posts over a certain period in time, we aim: (1) at capturing those sub-periods during which a user’s mood deviates from their baseline mood – a post-level sequential classification task. We then build on this task, by leveraging it to further help us assess: (2) the risk level the user is at – a user-level classification task & a continuation of the 2019 Shared Task. Thus, the task consists of the two subtasks: (1) the main task of identifying mood changes in users’ posts over time and (2) the auxiliary task of showing how (1) helps us assess the risk level of a user.
Data and processing environment
Social media data annotated for the purposes of the above tasks will be made available in a secure environment and all processing and model development will happen in the secure environment. This is because the data pertains to vulnerable individuals and we endeavour to protect them as much as possible. Task participants will be allocated credits in the secure environment and will be provided with the necessary development tools and libraries they request upon registering interest. Finally task participants will need to sign data use agreements and abide by ethical practice. In addition to the shared task, participants will be invited (and strongly encouraged) to participate in an online Hackathon scheduled for March 21-22. The aim of the Hackathon will be to try out the secure environment with a subset of the data prior to the shared task and permit some live interaction with the organising team in terms of their experience around the task. The organising team will also be providing baselines for the primary task (1) based on existing work.
Timeline
17 Feb – Call goes out inviting Expression of Interest
28 Feb – Deadline for teams to register interest (link)
7 Mar – Team selection announcement.
21 & 22 Mar – Hackathon- subset of training data available
1 Apr – Availability of training data
25 Apr – Availability of test data
1 May – System submissions due
5 May – Results announced
10 May – System description papers due
15 May – Acceptance notification
20 May – Camera ready
Quick Links
Invitation to Participate – Expression of Interest: https://docs.google.com/forms/d/e/1FAIpQLSfGGGjS3ZbBbG0eITF1hum-EdEzyKR2...
Email Organizers: clpsych22-shared-task-organizers [at] googlegroups.com
You can find information for last year’s shared task (CLPsych 2021) here (https://seanmacavaney.github.io/clpsych2021-shared-task/), and papers and presentations from the workshop here (https://clpsych.org/workshop-2021/#schedule).