Consideration of Factuality Analysis of Disease Symptoms for Visualization of Disease Symptoms based on Twitter

Abstract

In several studies, a method to detect the spread of infections such as influenza or to analyze its factuality from posted information on social media have been proposed. However, little studies are available on diseases and symptoms other than infectious diseases. Our study focuses on Twitter. We aim to propose a method to collect and analyze when, where, and what kinds of diseases and symptoms are tweeted, regardless of whether the diseases are infections or not, and to construct a system to visualize them by region and time series. In this paper, we consider a method to determine the factual status for disease symptoms using a typical binary classification model and examine various features in tweets with user’s disease symptoms. From experiments for 14 disease symptoms, we confirmed that a multilayer perceptron model with word2vec and a dictionary of Japanese functional expressions obtained the precision of 77%.