Automated learning of everyday patients’ language for medical blogs analytics

Proceedings of Recent Advances in Natural Language Processing, pages 640–648, Hissar, Bulgaria, 7-13 September 2013. 1 Automated Learning of Everyday Patients’ Language for Medical Blogs Analytics Giovanni Stilo Dipartimento di Informatica stilo@di.unirom a1.it Moreno De Vincenzi Dipartimento di Informatica devincenzi@di.u niroma1.it Alberto E. Tozzi Ospedale Pediatrico Bambin Gesù albertoeugeniot ozzi@opbg.net Paola Velardi Dipartimento di Informatica velardi@di.unir oma1.it Abstract Analyzing how people discuss about health-related topics on dedicated forums and social networks such as Twitter, can provide valuable insight for syndromic surveillance and to predict disease outbreaks. In this paper we present a minimally trained algorithm to learn associations between technical and everyday language terms, based on pattern generalization and complete linkage clustering, and we then assess its utility on a case study of five common syndromes for surveillance purposes. 1. Introduction Infodemiology is defined as “the science of distribution and determinants of information in an electronic medium, specifically the Internet, with the ultimate aim to inform public health and public policy” (Eysenbach, 2006). A seminal work in this area is (Ginsberg et al., 2009), in which the level of influenza in the U.S. is estimated using the relative frequency of search queries related to influenza-like illness. Similarly, in (Althouse et al., 2011), the authors demonstrate that query search volumes associated to Dengue fever can predict the incidence of Dengue. Another recent study (Xu et al., 2011) analyses the problem of predicting the tendency of hand-foot-and-mouth disease (HFMD), clustering HFMD-related search queries, medical pages and news reports. Query search volumes are estimated using Google Trends (GT) 1 or Google Flu, however, forums and micro-blogs (like Twitter) appear to be a better source of information, since keywords occur in contexts. Contexts make it possible to use text mining techniques for sense disambiguation, topic filtering and mood analysis (Berendt, 2011; Corley, 2009; Von Etter et al., 2010; Cohen and Hersh, 2005; Paul and 1 http://www.google.com/trends/ Dredze, 2011). Among the others, the problem of tracing patient’s naïve medical terminology is a very crucial one (Dahm, 2011; Molina Healthcare, 2004). Consider the following striking difference in the usage of terms describing the same health conditions, the first by a clinician, the second by a patient: “Clinicians should maintain a high index of suspicion for this diagnosis in patients presenting with influenza-like symptoms that progress quickly to respiratory distress and extensive pulmonary involvement.”2 “For the past 3 days I have had a stuffy, runny nose, congested chest, fever, sore ears and throat and burning eyes. I’ve been taking cold and flu medication, and it doesn't help”3. Clearly, the patient’s symptoms should induce “a high index of suspicion”, but for an automated system to capture a similarity between the two symptom descriptions is not obvious. Being able to understand the way people talk about their health conditions in “peer to peer” communications is crucial for an effective monitoring of health- related behaviors based on social data. In this paper we present a minimally supervised algorithm to learn patient’s jargon and we apply it to the analysis of …

Content truncated to 500 words