Waldo: Automated discovery of adverse events from unstructured self reports.

ABSTRACT: Adverse event (AE) detection is labor-intensive and costly given the task is to find rare events. Automated solutions to enhance efficiency, reduce costs, and capture unnoticed safety signals are needed. To develop and evaluate an automated machine learning tool, "Waldo," for AE detection from unstructured social media text data, specifically targeting consumer health products that lack traditional post-market surveillance channels. We tested three models - (i) N-gram model, (ii) BERT (Bidirectional Encoder Representations from Transformers), and (iii) RoBERTa (Robustly optimized BERT approach) - trained on 10,000 previously published unstructured reports on cannabis-derived products (CDPs) annotated by humans for the presence of adverse events to determine the best-performing AE detection

SUBMITTER: Desai KS 

PROVIDER: S-EPMC12483200 | biostudies-literature | 2025 Sep

REPOSITORIES: biostudies-literature