Cross Documents Concept Augmentation

Abstract

The current paper proposes a strategy for exploring and integrating related information extracted from unstructured documents with different degree of confidence, standardization and representation. The strategy was instantiated on the medical domain and designed for the English language. The goal of the proposed strategy was of augmenting the therapeutic information from patient leaflets with information extracted from clinical records. The approach proved to be a sound one as the information from the clinical records aligns with the information in the standardized sources. It confirmed the assumption that we can derive drug repositioning from clinical records and thus augmenting the existing medical knowledge. The reported metrics <95.14% precision, 83.3% recall>for patient leaflets and <94.07% precision, 87.27% recall >for EHRs measured for the concept extraction strategy, further support a good performance for the entities correlation approach. The degree of correlation between the extracted information from the two data sources reported as matches is of 85%.