Personal Health Mention Identification from Tweets Using Convolutional Neural Network
Yitu Wang, Xiaoli Li, Daniel Y. W. Mo · 2020
The past decade witnesses the unprecedent growth of social media users worldwide. People express health related outcomes, information, and views on social media platforms. This provides many opportunities to utilize the data source for health monitoring and surveillance, and digital epidemiology in real time. Personal health mention (PHM) is among one of the critical tasks for such purpose. It tries to identify whether a person's health condition is mentioned in a sentence. However, social media texts contain noises, many creative and novel phrases, sarcastic Emoji expressions, and misspellings. This poses challenges to detect PHM from social media text. This paper explores the PHM identification task for six diseases from twitter using convolutional neural network (CNN). Specifically, word embeddings are used to encode the twitter text. Then they are fed into CNN structure to train the classifier for PHM identification. We also explore how the performance of different methods are affected by data imbalance issue and training sample size.