Drug Drug Interaction Extraction from Chinese Biomedical Literature Using Distant Supervision

摘要

The field of pharmacovigilance has attracted widespread attention due to the increasing impact of drug safety incidents. In this paper, we try to extract Drug Drug Interactions (DDIs) from Chinese biomedical literature. In addition, we used a variety of biomedical resources to develop the first Chinese DDIs database with the help of expert annotations. Based on this database, we applied distant supervision method to extract DDIs from 11,319 biomedical sentences. In order to classify the relationship instances, we extract feature based on the Bidirectional Encoder Representation from Transformers (BERT) model, combine the attention mechanism to select effective instances, and provide drug descriptions to supplement background knowledge. At last. Our method achieves an F-score of 0.732, which is better than the traditional method. Furthermore, we analyze the false negatives in our results.