Learning to Extract Relations from MEDLINE

Information in text form remains a greatly underutilized resource in biomedical applications. We have begun a research effort aimed at learning routines for automatically mapping information from biomedical text sources, such as MEDLINE, into structured representations, such as knowledge bases. We describe our application, two learning methods that we have applied to this task, and our initial experiments in learning such information-extraction routines. We also present an approach to decreasing the cost of learning information-extraction routines by learning from "weakly" labeled training data.