Evaluating Semantic Similarity for Adverse Drug Event Narratives.

We propose a method to evaluate adverse drug event (ADE) narratives using biomedical semantic similarity measures. Automated drug surveillance systems have used social media as a prime resource to detect ADEs. However, the problem of language usage over social media has been a challenge in evaluating the performance of such systems. We address this key issue by using semantic similarity measures and the biomedical vocabularies from the Unified Medical Language System. This is important in comparing results of social media driven approaches against standard reference documents from regulatory agencies.