A clustering based methodology to support the translation of medical specifications to software models

摘要

In this paper we propose a methodology to reduce the complexity to realize a software validation model, starting from medical specifications written in Italian natural language text. In order to obtain an automatic validation system it is necessary to manually translate the specification documents into software models. This task is long, tedious and error prone, due to the manual effort needed. To speed up this process and to reduce the errors that can occur, an important boost can be obtained from the grouping of the conformance rules belonging to the same pattern. Clustering algorithms can accomplish this task, but there is the need to know a priori the total cluster number, and this is not possible in this kind of problem. At this aim, we propose two innovative automatic cluster selection methodologies able to evaluate the optimal number of clusters, based on an iterative internal cluster measure evaluation. These approaches consider three different Vector Space Models (VSMs), two different clustering algorithms and the impact of the using the Principal Component Analysis technique. The experimental assessment has been performed on four different datasets extracted from the HL7 CDA R2 Italian language conformance rules specification documents, demonstrating the effectiveness of the proposed methodology. Finally, in order to compare the results of all possible configurations, we realized a non-parametric statistical analysis. The obtained results demonstrated the effectiveness of the proposed methodology for automatic cluster number selection. (C) 2018 Elsevier B.V. All rights reserved.