Improving Relationship Extraction from Clinical Notes by Sentence Classification

摘要

In recent years, there has been an increasing interest in automating knowledge extraction from biomedical text, including both clinical notes and published literature. Extracting relationships between concepts in a sentence is an essential step in many knowledge extraction tasks. Identifying whether a sentence contains any relationship between its concepts, can potentially improve all relationship extraction methods. Here we seek to evaluate the effectiveness of binary sentence classification on relationship extraction from clinical notes, as well as the effect of different classes of features on the same task. We use 2010 i2b2/VA shared task clinical notes corpus as the gold standard for evaluation. The sentence binary classification achieves 90.14% f-measure (91.42% precision and 88.9% recall), improving the relationship extraction by 2.19% f-measure.