Ontology-based trace retrieval

Abstract

In automated requirements trace retrieval, an ontology can be used as an intermediary artifact to identify relationships that would not be recognized by standard information retrieval techniques. However, ontologies must be carefully constructed to fit the needs of the project. In this paper we present a technique for incorporating information from general and domain-specific ontologies into the tracing process. Our approach applies the domain ontology at the phrase level and then uses a general ontology to augment simple term matching in order to deduce relationships between individual terms weighted according to the relative importance of the phrase in which they occur. The combined weights are used to compute the overall similarity between a source and target artifact in order to establish a candidate trace link. We experimentally evaluated our approach against the standard Vector Space Model (VSM) and show that a domain ontology combined with generalized ontology returned greatest improvements in trace accuracy.