@inproceedings{aragon-etal-2024-delving,
    title = "Delving into the Depths: Evaluating Depression Severity through {BDI}-biased Summaries",
    author = "Arag{\'o}n, Mario Ezra  and
      Parapar, Javier  and
      Losada, David E.",
    editor = "Yates, Andrew  and
      Desmet, Bart  and
      Prud{'}hommeaux, Emily  and
      Zirikly, Ayah  and
      Bedrick, Steven  and
      MacAvaney, Sean  and
      Bar, Kfir  and
      Ireland, Molly  and
      Ophir, Yaakov",
    booktitle = "Proceedings of the 9th Workshop on Computational Linguistics and Clinical Psychology (CLPsych 2024)",
    month = mar,
    year = "2024",
    address = "St. Julians, Malta",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-emnlp/2024.clpsych-1.2/",
    pages = "12--22",
    abstract = "Depression is a global concern suffered by millions of people, significantly impacting their thoughts and behavior. Over the years, heightened awareness, spurred by health campaigns and other initiatives, has driven the study of this disorder using data collected from social media platforms. In our research, we aim to gauge the severity of symptoms related to depression among social media users. The ultimate goal is to estimate the user{'}s responses to a well-known standardized psychological questionnaire, the Beck Depression Inventory-II (BDI). This is a 21-question multiple-choice self-report inventory that covers multiple topics about how the subject has been feeling. Mining users' social media interactions and understanding psychological states represents a challenging goal. To that end, we present here an approach based on search and summarization that extracts multiple BDI-biased summaries from the thread of users' publications. We also leverage a robust large language model to estimate the potential answer for each BDI item. Our method involves several steps. First, we employ a search strategy based on sentence similarity to obtain pertinent extracts related to each topic in the BDI questionnaire. Next, we compile summaries of the content of these groups of extracts. Last, we exploit chatGPT to respond to the 21 BDI questions, using the summaries as contextual information in the prompt. Our model has undergone rigorous evaluation across various depression datasets, yielding encouraging results. The experimental report includes a comparison against an assessment done by expert humans and competes favorably with state-of-the-art methods."
}