A computational semantics system for detecting drug reactions and patient outcomes in personal health messages

Case studies are a standard approach to medicine. A physician needs the outcomes of a drug, situationally relevant to a particular patient. We propose a system for patient outcomes utilizing computational semantics, which effectively digests message groups. Filtering identifies personal comments, by eliminating useless messages. Clustering groups similar topics from different messages, by statistical overlap with specified terms. Summarizing labels the clusters so content can be quickly digested. We implemented a prototype system with these functions for mining health messages. Our methods do not require extensive training or dictionaries, while enabling users to specify custom topics for digesting. This system has been tested with sample messages from our unique dataset from Yahoo! Groups, containing 12M personal messages from 27K public groups in Health and Wellness. Evaluated results show high quality of labeled clustering, promising an effective automatic system for discovering useful information across large volumes of health information