Workshop Proceedings of the 19th International AAAI Conference on Web and Social Media
Workshop: #SMM4H-HeaRD 2025: Joint 10th Social Media Mining for Health and Health Real-World Data Workshop and Shared Tasks
DOI: 10.36190/2025.68This paper outlines techniques for predicting insomnia based on established rules and definitions. It delves into textual extraction that supports the end insomnia predictive task for evidence-based reasoning using Large Language Models (LLMs) as part of Shared Task, Task 4-SMM4H'25. We have utilised two LLMS - Gemma-2-2b-it and Meta-Llama-3-8b-Instruct-AWQ to identify patients with sleep disorders by extracting the text and approaching the problem reversely. Based on the sleep-related text extracted, we evaluate the rules and definitions and detect the insomnia disorder using the chosen LLMs. The approach showed an above-average mean and median F1 score of 0.8913 for Insomnia prediction and an above-average median F1 score of 0.6954 for identifying if the rules and definitions satisfy the insomnia prediction. Even though the other two modules relied on text extraction for end prediction, the text extraction phase had a lower mean and median F1 score of 0.4088.