We explore the contribution of bi-directional Long Short-Term Memory (bi-LSTM) units to Conditional Random Fields (CRFs) for extracting adverse drug reactions (ADRs) and ADR-related information (e.g., drug classes, severities). We utilize only real-valued word vectors as input. On the 2017 Text Analysis Conference (TAC) ADR dataset, we achieve a competitive micro-averaged F1-measure of 0.801, without benefiting from any hand-crafted features or external resources.
Learning Objective 1: Systematically analyzed three different methods for adverse drug reactions (ADRs) extraction.
Learning Objective 2: Proposed a competitive trainable system that does not rely on hand-crafted features and external resources.
Carson Tao (Presenter)
University at Albany, SUNY
Michele Filannino, George Mason University
Ozlem Uzuner, George Mason University