Solving the ADE Extraction Problem
This paper proposes a joint model for entity recognition and Adverse Drug Event (ADE) extraction in biomedical text, inspired by Machine Reading Comprehension (MRC) literature. By loosening the constraints on ADE extraction, the model outperforms state-of-the-art techniques by a large margin. The entity recognition sub-system also outperforms current best practices. By using an attention mechanism to facilitate intra-sequence interaction, we could replace SDP-based methods used in prior work.