New ACSS publication: Multilingual and Monolingual Adverse Event Classification using Transformers

by : Olivier Caron in : Workshop Proceedings of the 19th International AAAI Conference on Web and Social Media, 23 June, 2025, Copenhagen, Denmark.

Abstract : We participated in two shared tasks at the 2025 Social Media Mining for Health and Health Real-World Data (#SMM4H-HeaRD) workshop, focused on identifying adverse drug events (ADEs) from social media. For Task 1 (multilingual ADE classification), we fine-tuned an XLM-RoBERTa-Large model on English, French, German, and Russian datasets, achieving a macro F1-score of 0.633 on the test set. For Task 6 (Shingles vaccine adverse event and failure detection on Reddit), we used an ensemble of five DeBERTa-V3-Base models trained via 5-fold cross-validation, which reached an F1-score of 0.951. In both tasks, our systems outperformed the average and median performance of participating teams, confirming the competitiveness of our approach. These results demonstrate the effectiveness of transformer-based models for multilingual and noisy health-related social media classification tasks, even without additional text preprocessing.