Is There an Interplay Between Library Usage and Repository Features?
João Victor Esteves, Daniel Coutinho, Marcelo Schots, Igor Machado Coelho · 2019
The advent of open source has changed the way developers reuse software. The availability of libraries and their corresponding source code in public software repositories enables new forms of analyzing project aspects that can provide clues on their stability and maintainability. However, the literature lacks studies aiming to identify and understand whether and which repository features may correlate with the likeliness of usage of a library. In this sense, we present a factorial experiment using three different regression models - Multiple Linear Regression, Random Forest, and Neural Networks -, aiming at analyzing whether there is a correlation between library usage and a set of features extracted from release management and version control repositories. The results allowed to map features with positive learning impact, such as the number of stars, pull requests, and number of downloads, as well as features that contributed much less to the models (e.g., the repository size). Although the impact level of each feature varied from model to model, we also noticed from the analysis of regression results that the models were capable of achieving higher accuracy when considering only a subset of features.