Exploratory Analysis of Developer Sentiment On Open Source Projects

Abstract

Issue-tracking platforms such as Jira and Bugzilla have become essential in large- scale software development. Prominent organizations in the Open Source Software (OSS) landscape, such as Apache and Mozilla, make heavy use of these platforms and document their software development process through online repositories that utilize version control systems (VCS) like Git. Artifacts gathered from these sources contain natural language data that can be used to answer important questions relating to the nature of the software produced and the sentiment of the developers. The commit fre- quency and working time of the developers can be correlated to the sentiment shown through the commit messages. Moreover, the sentiment of issue comments might differ significantly based on the type (i.e., bug or non-bug) or severity. In this regard, we utilized a modern machine learning-based approach through fine-tuning seBERT, a BERT model pre-trained on software development data, to classify sentiment and provide answers to these questions. We used an existing data set, 20-MAD, to test these hypotheses and provide the results. We found that high committer frequency is associated with a higher proportion of negative sentiments compared to low and medium frequencies, while the part of the day developers work in has minimal effect on measured sentiment. We also observed that the severity of an issue significantly influences the sentiment expressed in issue comments and issues classified as bugs have a higher negative sentiment frequency compared to other issue types combined.