Building a Patient-based Ontology for User-written Web Messages

Proceedings of Recent Advances in Natural Language Processing, pages 758–763, Hissar, Bulgaria, 12-14 September 2011. Building a Patient-based Ontology for User-written Web Messages Marina Sokolova Faculty of Medicine, University of Ottawa and Electronic Health Information Lab, CHEO Research Institute sokolova@uottawa.ca David Schramm Faculty of Medicine, University of Ottawa and Children’s Hospital of Eastern Ontario The Ottawa Hospital dschramm@toh.on.ca Abstract We introduce an ontology that is represen- tative of health discussions and vocabulary used by the general public. The ontology structure is built upon general categories of information that patients use when de- scribing their health in clinical encoun- ters. The pilot study shows that the general structure makes the ontology useful in text mining of social networking web sites. 1 Introduction Recent studies have shown that public health surveillance benefits from information posted by users on the Web (Carneiro and Mylonakis, 2009; Ginsberg et al, 2008). Health-related mes- sages can be found on Web forums hosting so- cial networks (e.g., www.PatientsLikeMe. com) or individual blogs (e.g., http://www. jackslemonade.com). For medical professionals, the user-written health information assists in prediction of public attitude towards health policies. In user messages, patient-based information prevails over biomed- ical information. Patient-based information is brought forth when a user views himself as a po- tential or real patient of a health care provider. This information reveals details of one’s health that are usually discussed during visits to a health care provider. Patient-based information is often identified as evidence-based, whereas the biomed- ical information is viewed as knowledge-based (Hersh, 2009). Development of social media has prompted refocusing of text analysis from biomedical to patient-based health information mining. Several academic groups actively work on health infor- mation studies (Angelova, 2010; Chapman, 2010; Chanlekha and Collier, 2010). These groups work on methods for the analysis of academic and pro- fessional articles in medical journals and tradi- tional news media, as well as hospital documen- tation. At present, user-written health information is the subject of studies by data mining where the analysis primarily relies on statistical methods (Lampos and Christianini, 2010) and public health informatics which usually addresses specific ques- tions, e.g., injury discussions by military service- men (Konovalov et al, 2010). A prevalent trend in health-related text analy- sis is to solve a particular task which is closely associated with a particular data source, e.g., iden- tifying involuntary childlessness terminology on a dedicated web site (Himmel et al., 2009) or find- ing new terms used on a patient social networking site (Doing-Harris and Zeng-Treiler, 2011). The specific focus makes the accumulated knowledge inherently individualized towards the task and the data. It permits high accuracy on the original data, whereas shifting to other data sets is likely to ex- perience performance set-back. Our goal is to build a patient-based resource organized as an ontology, a repository of health- related terms assigned into a hierarchical structure of semantic categories. The general categories are durable and able to withstand the rapidly evolving environment of the Web. In an empirical setting, we show that …

Content truncated to 500 words