Predicting escalations of medical queries based on web page structure and content

Logs of users' searches on Web health topics can exhibit signs of escalation of medical concerns, where initial queries about common symptoms are followed by queries about serious, rare illnesses. We present an effort to predict such escalations based on the structure and content of pages encountered during medical search sessions. We construct and then characterize the performance of classifiers that predict whether an escalation will occur after the access of a page. Our findings have implications for ranking algorithms and the design of search interfaces.