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: 10.36190/2025.77

Published: 2025-06-05
BrynMawrNLP @ SMM4H-HeaRD 2025: Stacking RoBERTA with Naive Bayes to Identify Personal Shingles Vaccine Reactions on Reddit
Joon Luther, Adam Poliak

We present the BrynMawr NLP System for the 10th Social Media Mining for Health Research and Applications Workshop (SMM4H 2025), addressing Task 6 - distinguishing Reddit posts that contain personal mentions of adverse reactions to herpes zoster (shingles) vaccines from other vaccine-related discussions. Our system is based on a simple logistic regression model trained on the predictions from a fine-tuned RoBERTa model and Naive Bayes model. We explore stacking a fine-tuned RoBERTa model with traditional non-neural models to improve upon just fine-tuning a pre-trained RoBERTA model. Our experiments demonstrate that statistical, non-neural models perform well on this dataset but their performance do not compare with those of a fine-tuned Transformer.