Visualization of the occurrence trend of infectious diseases using Twitter

Visualization of the occurrence trend of infectious diseases using Twitter Ryusei Matsumoto, Minoru Yoshida, Kazuyuki Matsumoto, Hironobu Matsuda, Kenji Kita Institute of Technology and Science, University of Tokushima 2-1, Minami-josanjima, Tokushima, 770-8506, Japan c501306048@tokushima-u.ac.jp, mino@is.tokushima-u.ac.jp, matumoto@is.tokushima-u.ac.jp, c501637026@tokushima-u.ac.jp, kita@is.tokushima-u.ac.jp Abstract We propose a system for visualizing the epidemics of infectious diseases. We apply factuality analysis both to disease detection and location estimation for accurate visualization. We tested our methods for several infectious diseases, and show that our method performs well on various diseases. Keywords: Twitter, disease surveillance, location estimation 1. Introduction Information related to human life such as illness and infec- tious diseases is important for many people. However, peo- ple have limited kinds of sources of information on diseases such as TV news and newspapers (including Internet news.) On the other hand, SNS including Twitter is a promising source of another type of information about diseases. It has a highly real-time nature and can obtain a large amount of information about the event occurring just now. The final goal of this research is to provide users with real-time in- fection maps automatically created from Twitter posts. To this end, we propose a system to obtain how many people are infected by the given disease in each area. The pro- posed system extracts the tweets that contain the disease name, and estimates the location of the user. The system performs factuality analysis of mentions both for infection events and user locations for more accurate estimation. Fi- nally, an infectious disease map is created using the result of the location estimation. 2. Existing Research Several researches that combine Twitter and illness have been done so far (Charles-Smith et al., 2015). For exam- ple, the system by Aramaki et al. (Aramaki et al., 2011; Iso et al., 2016; Kanouchi et al., 2015; Kitagawa et al., 2015) extracts tweets related to influenza posted on Twit- ter, conducted factuality analysis using SVMs on extracted tweets for detecting epidemic influenza. They also provide the ”influ kun” service1 which extracts tweets related to in- fluenza from Twitter and provides a epidemic map of in- fluenza overlaid on a map of Japan. Other existing research on disease surveillance with location estimation used sim- ple geocoding (Li and Cardie:, 2013), geotagged tweets (Sadilek et al., 2012) and user profiles (Dredze et al., 2013; Paul and Dredze, 2011). Our system is different from this service in that it is aimed at various kinds of diseases other than influenza, and estimate users location even if the user location is not explicitly provided in the form of geotag or user profiles. There have also been many researches on estimating user locations from tweets (contents) (Chandra et al., 2011; 1 http://mednlp.jp/influ map/ Cheng et al., 2010; Han et al., 2012; Sugiya et al., 2013). Most of the research use the geotagged tweets as training data and estimate the positions of users/tweets based on the word-geotag collocations. The goal of those previous location estimation researches is to estimate the locations vaguely with best-effort approaches. …

Content truncated to 500 words