This paper describes the contributions of the BLUE team in the CLPsych 2025 Shared Task on Capturing Mental Health Dynamics from Social Media Timelines. We participate in all tasks with three submissions, for which we use two sets of approaches: an unsupervised approach using prompting of various large language models (LLM) with no fine-tuning for this task or domain, and a supervised approach based on several lightweight machine learning models trained to classify sentences for evidence extraction, based on an augmented training dataset sourced from public psychological questionnaires. We obtain the best results for summarization Tasks B and C in terms of consistency, and the best F1 score in Task A.2.

This paper describes the Unibuc Archaeology team work for CLPsych’s 2024 Shared Task that involved finding evidence within the text supporting the assigned suicide risk level. Two types of evidence were required: highlights (extracting relevant spans within the text) and summaries (aggregating evidence into a synthesis). Our work focuses on evaluating Large Language Models (LLM) as opposed to an alternative method that is much more memory and resource efficient. The first approach employs an LLM that is used for generating the summaries and is guided to provide sequences of text indicating suicidal tendencies through a processing chain for highlights. The second approach involves implementing a good old-fashioned machine learning tf-idf with a logistic regression classifier, whose representative features we use to extract relevant highlights.