Attention-Based Multi-Task Learning in Pharmacovigilance

Pharmacovigilance (PV) is the process of monitoring and assessing adverse events (AEs). Currently, the primary method of analysis in PV is manual inspection by individual case managers. However, this is largely unsustainable due to the increased volume of cases over the past few years. Since AE processing involves several sub-tasks, such as annotation and classification, our paper explores a novel solution to PV. In this paper, we propose a multi-task learning (MTL) model, where the hidden layers are shared, to jointly learn the tasks (Named Entity Recognition (NER), Classification). The results of our paper demonstrate that MTL is able to outperform our baseline classification model and equal the baseline model for annotation/NER.