Evaluation-Oriented Analysis of the Traceability Methods in Software Systems
TL;DR
This paper focuses on evaluation data from the available literature that provide an insight into the comparative performance of the approaches and the basic metrics observed were: standard, Recall, and F-score, and specific ones.
Abstract
: Nowadays, software projects have become extremely large-scale, and their development requires the application of numerous standard and innovative technologies. Hence, the possibility of embedding and maintaining information about the mutual dependencies of the components in the software system becomes one of the key and necessary activities during development. This traceability feature enables the monitoring of the artifact connections and helps in preserving the system integrity. First, this paper briefly introduces the concept itself and the three most relevant groups of traceability methods: IR-based, ML-based, and graph-based. Then, the paper focuses on evaluation data from the available literature that provide an insight into the comparative performance of the approaches. The basic metrics observed were: standard (Precision, Recall, and F-score), and specific ones. Different evaluation setups and configurations are also considered. Although some papers indicate that ML-based methods can outperform IR-based ones, no general conclusion can be drawn because of quite different datasets and evaluation environments.