Healthy or harmful? polarity analysis applied to biomedical entity relationships

In this paper, we investigate how to automatically identify the polarity of relationships between food and disease in biomedical text. In particular, we first analyze the characteristic and challenging of relation polarity analysis, and then propose a general approach, which utilizes background knowledge in terms of word-class association, and refines this information by using domain-specific training data. In addition, we propose several novel learning features. Experimental results on real world datasets show that the proposed approach is effective.