Extracting Meaningful Correlations among Heterogeneous Datasets for Medical Question Answering with Domain Knowledge

Abstract

A question answering system (QAS) merely built on a predefined medical knowledge base experiences difficulties in providing suitable answers for expert users to make medical and healthcare decisions. This study proposes a comprehensive method of extracting meaningful correlations among heterogeneous datasets using a semantic analysis with domain knowledge and accordingly provide flexible answers to decision support (ATDS) in a medical QAS (MQAS). First, the potential value of the heterogeneous datasets from medical information systems is examined for building ATDS. Second, an extraction algorithm for constructing a term relational network from the questions is proposed. Then, a correlation construction method for integrating the datasets into the MQAS using domain knowledge is proposed. Finally, a novel algorithm for constructing ATDS on the basis of questions and datasets is established. Experimental results indicate that utilizing external medical domain knowledge in analyzing correlations among the datasets outperforms existing algorithms that only involved with the datasets.