Condition Unknown: Predicting Patients' Health Conditions in an Online Health Community

Online health communities rely on information about their users to provide services to members. We partner with the online health community CaringBridge.org to infer the health condition that users are discussing from their early writing on the site. We utilize the self-reported health condition data that is provided by users to train machine learning classifiers to predict the health condition of non-reporting users. An analysis of the classifier's errors reveals that users frequently discuss multiple health conditions. We present models with explainable features, enabling us to extract words for the enrichment of consumer health vocabularies and to support future designs connecting patients.