Detecting Medication Mentions in Social Media Data Using Large Language Models
Lopez-Garcia, G.; Xu, D.; Gonzalez-Hernandez, G.
Show abstract
The automatic extraction of medication mentions from social media data is critical for pharmacovigilance and public health monitoring. In this study, we present an end-to-end generative approach based on instruction-tuned large language models (LLMs) for medication mention extraction from Twitter. Reformulating the task as a text-to-text generation problem, our models achieve state-of-the-art results on both fine-grained span extraction and coarse-grained tweet-level classification, surpassing traditional sequence labeling baselines and previous best-performing systems. We demonstrate that fine-tuning Flan-T5 models enables efficient and accurate extraction while simplifying the architecture by eliminating complex multi-stage pipelines. Additionally, we show that lexicon-based filtering further improves performance by reducing false positives. Our models are publicly available, providing high-performing and efficient tools for large-scale pharmacological analysis of social media data.
Matching journals
The top 6 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
- Zero Shot Health Trajectory Prediction Using Transformer 95%
- FedWeight: Mitigating Covariate Shift of Federated Learning on Electronic Health Records Data through Patients Re-weighting 93%
- Clinical Knowledge Extraction via Sparse Embedding Regression (KESER) with Multi-Center Large Scale Electronic Health Record Data 93%
Similar papers in this journal
Similar papers in this journal
- Natural Language Processing for Automated Annotation of Medication Mentions in Primary Care Visit Conversations 94%
- Using indication embeddings to represent patient health for drug safety studies 94%
- A Study of Calibration as a Measurement of Trustworthiness of Large Language Models in Biomedical Research 94%
Similar papers in this journal
- Building a Best-in-Class De-identification Tool for Electronic Medical Records Through Ensemble Learning 95%
- Inferring global-scale temporal latent topics from news reports to predict public health interventions for COVID-19 95%
- Federated Learning for multi-omics: a performance evaluation in Parkinson's disease 92%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.