Call for Participation: eRisk Lab @ CLEF 2025 - Corpora

6 Dec 2024


      -------- Original Message --------
Subject: Call for Participation: eRisk Lab @ CLEF 2025
Date: 2024-12-05 19:50
 From: ACL Announcements announcements@aclweb.org
To: Announcements announcements@aclweb.org
Call for Participation: eRisk Lab @ CLEF 2025
Are you passionate about leveraging AI for societal good? Join us for 
eRisk 2025, the ninth edition this lab at CLEF, where we delve into the 
methodologies and applications of early risk detection on the Internet. 
Our mission is to foster interdisciplinary research that addresses 
critical health and safety challenges, from identifying signs of 
depression to preventing online harm.
Tasks for eRisk 2025 (More info at https://erisk.irlab.org/ )
Task 1: Search for Symptoms of Depression
- Objective: Rank sentences from user writings by relevance to the 21 
symptoms of the BDI-II questionnaire.
- Highlights:
   - Use a TREC-formatted dataset with human-assessed relevance 
judgments.
   - Generate rankings for symptoms with evaluation via metrics like MAP 
and nDCG.
   - Create a valuable annotated corpus with broad applications beyond 
this task.
   - This is the third edition of the task: two years of training data.
Task 2: Contextualized Early Detection of Depression *(New in 2025)*
- Objective: Analyze full conversational contexts to detect early signs 
of depression.
- Highlights:
- Evaluate sequential user interactions for a holistic view of 
conversational dynamics.
    - Train on isolated writings and test in real-world-like scenarios 
with chronologically ordered conversations.
   - Metrics include accuracy and timeliness, measured via ERDE and 
similar frameworks.
    - This is the first edition of the contextualized tasks: three year 
of un-contextualized training data.
Pilot Task: Conversational Depression Detection via LLMs (New in 2025, 
Interactive Task)
- Objective: Engage with LLM personas to identify depressive symptoms 
based on conversational exchanges.
- Highlights:
   - No training data provided—use creative and unsupervised approaches.
   - Collaborate in a limited-message dialogue setting, simulating 
real-world conditions.
   - Push the boundaries of AI-human interaction for mental health 
applications: are we able to accurately reproduce personas?
   -This is a pilot task. Participants will need to book a slot to 
interact with the LLM personas: register before the slots are gone!
Key Dates
- Dataset Release:
     -T1: 1st December 2024 for training collections and test dataset
     -T2: 1st December 2024 for training and 5th February 2025 for 
beginning of test stage (server opens)
     -T3: 5th February 2025 for beginning of test stage (server opens for 
interacting with the LLM)
- Submission Deadlines:
     -T1: 1st April 2025 for submitting participants’ results to FTP
     -T2: 12th April 2025 end of test stage (server closes)
     -T3: 12th April 2025 end of test stage (server closes)
- CLEF 2025 Conference: 9-12 September 2025, Madrid, Spain.
How to Participate
1. Register: Sign up through the [CLEF 2025 Labs Registration 
site](https://clef2025-labs-registration.dei.unipd.it/)
2. Submit Agreements: Complete the user agreement form to access 
datasets.
3. Join the Community: Join our Google Groups 
https://groups.google.com/g/erisk-clef !
Lab co-chairs
Javier Parapar, Univ. A Coruña, Spain
Anxo Pérez, Univ. A Coruña, Spain
Xi Wang, Univ. Sheffield, United Kingdom
Fabio Crestani, Univ. Lugano, Switzerland
More Information
Visit the [eRisk website](https://erisk.irlab.org) for task details, 
datasets, and registration guidelines.