Flexible Concept Matching for Medical Information Retrieval

Extensive research efforts have been spent on concept-based medical information retrieval, relying on rich domain resources and tools. In most cases, the identified medical concepts in a query are either matched strictly against a document (i.e. concepts should be identical), or they are transformed into a bag of words. While the first approach has had limited success due to problems in concept identification, there is a serious loss in precision with the second approach. In this paper, we propose a flexible matching approach by considering a concept expression as a phrase, which can be matched strictly, or in some variant forms by allowing some flexibility in position and order. Such flexibility is naturally indicated by the great variations to express concepts in this area. In addition, we also incorporate synonym expressions in the matching process. Our experiments on several public test collections show that the flexible concept matching approach outperforms the traditional bag-of-words approach and the strict concept matching approach.