Collecting health related text from patient health writings

The Internet has been a huge resource for sharing and collecting information including health related information. Some health related information is written by patients (lay persons) discussing their experience about health problems and treatments. This paper introduces our initial work on providing physicians with clinically useful patient health writings. More specifically, the paper presented our experiments, as a part of the whole research work, on filtering health related text from patient health writings. We focused on selecting possible feature for classifying text from breast cancer mailing list into health and non health related text. Using KNN classification method, we experimented with various features, i.e. all terms, all terms except most frequently used terms, UMLS terms, health related UMLS terms, and health related UMLS semantic types. The experiments showed that UMLS terms extracted from the text is a good feature, compared to the other features.