A Systematic Literature Review of Multi-Label Learning in Software Engineering

In this article, we provide the first systematic literature review of the intersection of two research areas: Multi-Label Learning (MLL) and Software Engineering (SE). We refer to this intersection as MLL4SE. In recent years, MLL problems have increased in many applications and research areas because real-world datasets often have a multi-label nature. For multi-label data, simplifying the assumption of traditional classification approaches that an instance can only be associated with one class only leads to worse accuracy. Thus, a better match of methods and assumptions about the data is required. We identified 50 primary studies in our systematic literature review in the MLL4SE domain. Based on this review, we identified six main SE application domains where MLL has been applied. These domains include Software Requirement Engineering, Issue Tracking and Management, Community and Knowledge Management, API Usage and Management, Code Quality and Maintenance, and Mobile Application Development . We summarized the methods used and the data nature of the MLL4SE applications. Moreover, we separately provide taxonomies of future work directions from machine learning and SE perspectives. In general, we highlight current trends, research gaps, and shortcomings.