A Graph-boosted Framework for Adverse Drug Event Detection on Twitter

摘要

Detecting adverse drug events from Twitter is expected to reveal unreported side effects, thereby complementing current spontaneous reporting systems. However, existing studies usually only use word embeddings as the input for deep learning models, which ignores the structural information of sentences. In addition, deep learning models usually require a large number of cases for training, but the scale of annotated corpora that can be used for this task is limited. In order to solve the above problems, we propose a graph-boosted framework, that constructs the text into a graph structure. By using pre-trained graph embeddings and word embeddings for model training, our proposed framework provides richer semantic and structural information for prediction. The experimental results show that the proposed method can be used in different deep learning models and bring improvements when using the TwiMed corpus of different scales.