serguntsov/bert-base-uncased-drugsCom_raw-cls

Drug Review Sentiment Analysis with BERT

A fine-tuned BERT model for detecting positive/negative sentiment in pharmaceutical drug reviews with state-of-the-art performance. Developed as part of the DRUG_FEEDBACK_NLP project.

Model Overview

• Architecture: bert-base-uncased fine-tuned
• Input: Drug review text
• Output: Binary sentiment (0 = negative, 1 = positive)
• Threshold: Ratings >7 converted to positive (1)
• Max Sequence Length: 512 tokens
• GitHub Project: DRUG_FEEDBACK_NLP

Performance

Metric	Score
ROC-AUC	0.967
F1-Score	0.936
Precision	0.932
Recall	0.939