An Improved Approach to Traceability Recovery Based on Word Embeddings

摘要

Software traceability recovery, which reconstructs links between software artifacts, has become more and more vital to maintaining a software life cycle with the increase of software scale and complexity of software architecture. However, existing approaches mainly rely on information retrieval (IR) techniques. These methods are not very efficient at complex software artifacts which are mixed with multilingual texts, code snippets and proper nouns. Moreover, it is hard to predict new traceability links with existing approaches when requirements are changed or software functions are added, since these methods have not made the most of the final ranked lists. In this paper, we propose a novel approach WELR, based on word embeddings and learning to rank to recover traceability links. We use word embeddings to calculate semantic similarities between software artifacts and bring in query expansion and a weighting strategy during calculation. Different from other work, we leverage learning to rank to build prediction models for traceability links. We conducted experiments on five public datasets and took account of traceability links among different kinds of software artifacts. The results show that our method outperforms the state-of-the-art method that works under the same conditions.