CareLab at #SMM4H-HeaRD 2025: Insomnia Detection and Food Safety Event Extraction with Domain-Aware Transformers

2025-06-22

AI Summary

This study addresses two cross-domain information extraction tasks: insomnia symptom detection in clinical texts and food safety incident extraction in news articles. Methodologically, we propose a domain-aware Transformer framework built upon RoBERTa-style encoders, integrating three key innovations: (1) GPT-4–driven large language model (LLM) data augmentation, (2) task-adaptive input construction, and (3) domain-adaptive fine-tuning. These components collectively enhance domain-specific semantic modeling and few-shot generalization over generic models. Evaluated on SMM4H-HeaRD 2025 Task 5 Subtask 1 (food safety incident extraction), our approach achieves an F1 score of 0.958—ranking first—and attains state-of-the-art performance on the insomnia detection subtask. Our work empirically validates the efficacy of synergistic LLM augmentation and domain-informed encoding, establishing a reusable technical paradigm for low-resource information extraction in healthcare and public safety domains.

Technology Category

• Natural Language Processing: Information Extraction
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Application Domains: Humanities & Computational Social Science

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

This paper presents our system for the SMM4H-HeaRD 2025 shared tasks, specifically Task 4 (Subtasks 1, 2a, and 2b) and Task 5 (Subtasks 1 and 2). Task 4 focused on detecting mentions of insomnia in clinical notes, while Task 5 addressed the extraction of food safety events from news articles. We participated in all subtasks and report key findings across them, with particular emphasis on Task 5 Subtask 1, where our system achieved strong performance-securing first place with an F1 score of 0.958 on the test set. To attain this result, we employed encoder-based models (e.g., RoBERTa), alongside GPT-4 for data augmentation. This paper outlines our approach, including preprocessing, model architecture, and subtask-specific adaptations

Problem

• Detect insomnia mentions in clinical notes
• Extract food safety events from news articles
• Improve performance using domain-aware transformer models

Innovation

• Domain-aware transformers for insomnia detection
• RoBERTa and GPT-4 for data augmentation
• Preprocessing and model architecture adaptations