A Machine Learning Approach for Determining the Validity of Traceability Links

摘要

Traceability Link Recovery (TLR) is a fundamental software maintenance task in which links are established between related software artifacts of different types (e.g., source code, documentation, requirements specifications, etc.) within a system. Existing approaches to TLR often require a human to analyze a long list of potential links and distinguish valid links from invalid ones. Here we present an approach which bypasses this intermediate step and automatically classifies links as valid or invalid using a machine learning approach and features such as text retrieval (TR) rankings and query quality (QQ) metrics. We performed an evaluation on recovering traceability links in three software systems and the results show the potential of our approach, which achieved 95% accuracy on average using both types of features.