Trace-by-classification: A machine learning approach to generate trace links for frequently occurring software artifacts

Abstract

Over the past decade the traceability research community has focused upon developing and improving trace retrieval techniques in order to retrieve trace links between a source artifact, such as a requirement, and set of target artifacts, such as a set of java classes. In this Trace Challenge paper we present a previously published technique that uses machine learning to trace software artifacts that recur is similar forms across across multiple projects. Examples include quality concerns related to non-functional requirements such as security, performance, and usability; regulatory codes that are applied across multiple systems; and architectural-decisions that are found in many different solutions. The purpose of this paper is to release a publicly available TraceLab experiment including reusable and modifiable components as well as associated datasets, and to establish baseline results that would encourage further experimentation.