Identifying Potential Adverse Drug Events in Tweets Using Bootstrapped Lexicons

Adverse drug events (ADEs) are medical complications co-occurring with a period of drug usage. Identification of ADEs is a primary way of evaluating available quality of care. As more social media users begin discussing their drug experiences online, public data becomes available for researchers to expand existing electronic ADE reporting systems, though non-standard language inhibits ease of analysis. In this study, portions of a new corpus of approximately 160,000 tweets were used to create a lexicon-driven ADE detection system using semi-supervised, pattern-based bootstrapping. This method was able to identify misspellings, slang terms, and other non-standard language features of social media data to drive a competitive ADE detection system.