MLCQ

Abstract

Context Research on code smells accelerates and there are many studies that discuss them in the machine learning context. However, while data sets used by researchers vary in quality, all which we encountered share visible shortcomings---data sets are gathered from a rather small number of often outdated projects by single individuals whose professional experience is unknown. Aim This study aims to provide a new data set that addresses the aforementioned issues and, additionally, opens new research opportunities. Method We collaborate with professional software developers (including the code quest company behind the codebeat automated code review platform integrated with GitHub) to review code samples with respect to bad smells. We do not provide additional hints as to what do we mean by a given smell, because our goal is to extract professional developers' contemporary understanding of code smells instead of imposing thresholds from the legacy literature. We gather samples from active open source projects manually verified for industry-relevance and provide repository links and revisions. Records in our MLCQ data set contain the type of smell, its severity and the exact location in source code, but do not contain any source code metrics which can be calculated using various tools. To open new research opportunities, we provide results of an extensive survey of developers involved in the study including a wide range of details concerning their professional experience in software development and many other characteristics. This allows us to track each code review to the developer's background. To the best of our knowledge, this is a unique trait of the presented data set. Conclusions The MLCQ data set with nearly 15000 code samples was created by software developers with professional experience who reviewed industry-relevant, contemporary Java open source projects. We expect that this data set should stay relevant for a longer time than data sets that base on code released years ago and, additionally, will enable researchers to investigate the relationship between developers' background and code smells' perception.