Who caught a cold ? - Identifying the subject of a symptom

Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing, pages 1660–1670, Beijing, China, July 26-31, 2015. c⃝2015 Association for Computational Linguistics Who caught a cold? — Identifying the subject of a symptom Shin Kanouchi†, Mamoru Komachi†, Naoaki Okazaki‡, Eiji Aramaki§, and Hiroshi Ishikawa† †Tokyo Metropolitan University, {kanouchi-shin at ed., komachi at, ishikawh at}tmu.ac.jp ‡Tohoku University, okazaki at ecei.tohoku.ac.jp §Kyoto University, eiji.aramaki at gmail.com Abstract The development and proliferation of so- cial media services has led to the emer- gence of new approaches for surveying the population and addressing social issues. One popular application of social media data is health surveillance, e.g., predicting the outbreak of an epidemic by recogniz- ing diseases and symptoms from text mes- sages posted on social media platforms. In this paper, we propose a novel task that is crucial and generic from the viewpoint of health surveillance: estimating a sub- ject (carrier) of a disease or symptom men- tioned in a Japanese tweet. By designing an annotation guideline for labeling the subject of a disease/symptom in a tweet, we perform annotations on an existing cor- pus for public surveillance. In addition, we present a supervised approach for pre- dicting the subject of a disease/symptom. The results of our experiments demon- strate the impact of subject identification on the effective detection of an episode of a disease/symptom. Moreover, the results suggest that our task is independent of the type of disease/symptom. 1 Introduction Social media services, including Twitter and Face- book, provide opportunities for individuals to share their experiences, thoughts, and opinions. The wide use of social media services has led to the emergence of new approaches for survey- ing the population and addressing social issues. One popular application of social media data is flu surveillance, i.e., predicting the outbreak of in- fluenza epidemics by detecting mentions of flu in- fections on social media platforms (Culotta, 2010; Lampos and Cristianini, 2010; Aramaki et al., 2011; Paul and Dredze, 2011; Signorini et al., 2011; Collier, 2012; Dredze et al., 2013; Gesualdo et al., 2013; Stoové and Pedrana, 2014). Previous studies mainly relied on shallow tex- tual clues in Twitter posts in order to predict the number of flu infections, e.g., the number of oc- currences of specific keywords (such as “flu” or “influenza”) on Twitter. However, such a simple approach can lead to incorrect predictions. Bro- niatowski et al. (2013) argued that media atten- tion increases chatter, i.e., the number of tweets that mention the flu without the poster being ac- tually infected. Examples include, “I don’t wish the flu on anyone” and “A Harry Potter actor hos- pitalised after severe flu-like syndromes.” Lazer et al. (2014) reported large errors in Google Flu Trends (Carneiro and Mylonakis, 2009) on the ba- sis of a comparison with the proportion of doctor visits for influenza-like illnesses. Lamb et al. (2013) aimed to improve the ac- curacy of detecting mentions of flu infections. Their method trains a binary …

Content truncated to 500 words