Comparing Different Developer Behavior Recommendation Styles

Abstract

Research shows that one of the most effective ways software engineers discover useful developer behaviors, or tools and practices designed to help developers complete programming tasks, is through human-to-human recommendations from coworkers during work activities. However, due to the increasingly distributed nature of the software industry and development teams, opportunities for these peer interactions are in decline. To overcome the deprecation of peer interactions in software engineering, we explore the impact of several system-to-human recommendation systems, including the recently introduced suggested changes feature on GitHub which allows users to propose code changes to developers on contributions to repositories, to discover their impact on developer recommendations. In this work, we aim to study the effectiveness of suggested changes for recommending developer behaviors by performing a user study with professional software developers to compare static analysis tool recommendations from emails, pull requests, issues, and suggested changes. Our results provide insight into creating systems for recommendations between developers and design implications for improving automated recommendations to software engineers.