Towards ontology enrichment with treatment relations extracted from medical abstracts

Abstract

In this paper, we present the results of experiments identifying the drug treatment relation and drug treatment attributes like dosage, treatment frequency and duration from abstracts of medical publications using linguistic patterns. The approach uses an automatic linguistic pattern construction algorithm after the dataset has been semantically annotated. The automatically constructed patterns were able to identify treatment relations and their attributes with varying success. We observe that the simple (or naive) treatment patterns performs much better than the non-naïve treatment patterns in identifying sentences with drug treatment relationship in both cancer and non-cancer drug therapy domain. However the drug dosage, frequency and duration patterns performed much better in the identification of relationships in the cancer drug therapy domain than the non-cancer drug therapy domain. © Springer-Verlag Berlin Heidelberg 2006.