Drug-BERT : Pre-trained Language Model Specialized for Korean Drug Crime

We propose Drug-BERT, a specialized pre-trained language model designed for detecting drug-related content in the Korean language. Given the severity of the current drug issue in South Korea, effective responses are imperative. Focusing on the distinctive features of drug slang, this study seeks to improve the identification and classification of drug-related posts on social media platforms. Recent drug slangs are gathered and used to collect drug-related posts, and the collected data is used to train the language model. The designed pre-trained model is DRUG-BERT. The results show that fine-tuned DRUG-BERT outperforms that of the comparative models, achieving a 99.43% accuracy in classifying drug-relevant posts. Drug-BERT presents a promising solution for combatting drug-related activities, contributing to proactive measures against drug crimes in the Korean context.