Social media based adverse drug reaction detection

EasyChair Preprint 1383

Abstract

With the rapid development of the Internet, people are using social media as the main platform for knowledge sharing and emotional communication. Therefore, the detection of adverse drug reactions in social media will be an important way to pay attention to the current situation of people taking drugs. At the same time, traditional machine learning is difficult and weakly migrating when constructing features, and convolution neural networks (such as CNN) have the disadvantages of low efficiency and space insensitivity when constructing spatial information. Aiming at the above problems, this paper proposes a method based on capsule network and long-short memory neural network to detect adverse drug reaction events in social media based on general text processing features and biomedical proprietary features. We use the data which is the 2017 Third Social Media Mining for Health (SMM4H) shared task corpus. After the process of the corpus, marks the adverse drug reactions, and constructs distributed word vector features, part-of-speech tags, character-level vector features, and in every sentence, the drug name and the emotional word were input as the characteristics of the model. It solves the problem of the lack of spatial relationship between features and the low efficiency of the construction model in the classification process. The experimental results are compared with the previous advanced results, the F1 value is increased by 4.2%, which proves that the method is detecting the adverse drug reaction events of social media. Medium is effective and has good performance.

Keyphrases: Capsule Network, adverse drug reactions, biomedicine, social media

@booklet{EasyChair:1383,
  author    = {Zhu Xiaoxu and Lin Hongfei and Zeng Zeyuan},
  title     = {Social media based adverse drug reaction detection},
  howpublished = {EasyChair Preprint 1383},
  year      = {EasyChair, 2019}}