A supervised named entity recognition for information extraction from medical records

Abstract

— Named entity recognition is a widely used task to extract various kinds of information from unstructured text. Medical records, produced by hospitals every day contain huge amount of data about diseases, medications used in treatment and information about treatment success rate. There are a large number of systems used in information retrieval from medical documentation, but they are mostly used on documents written in English language. This paper contains the explanation of our approach to solving the problem of extracting disease and drug names from medical records written in Serbian language. Our approach uses statistical language models and can detect up to 80% of named entities, which is a good result given the very limited resources for Serbian language, which makes the process of detection much more difficult.