Improving GitHub Tag Recommender Systems Using Tag Hierarchies

Bachelor thesis (2022)

Authors

A.C. van der Rande Electrical Engineering, Mathematics and Computer Science

Contributors

M. Izadi Software Engineering - EEMCS (mentor)
A. van Deursen Software Technology (mentor)
G. Iosifidis Embedded Systems - EEMCS (graduation committee member)
Faculty
Electrical Engineering, Mathematics and Computer Science, Electrical Engineering, Mathematics and Computer Science
Copyright: © 2022 Arend van der Rande
More Info
expand_more

Published Date

22-06-2022

Language

English

Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Faculty

Electrical Engineering, Mathematics and Computer Science

Abstract

Programmers and software engineers often share code and one of the largest platforms on which this happens is GitHub, with an 87,58\% market share in the Source Code Management Category. One important part of sharing code is making sure that others who might be interested in it are also able to find it. One way to do that is by adding tags to a repository, which is a feature on GitHub only since 2017. However, many repositories do not have any assigned tags. This can be solved by automatically applying tags to repositories without any. This comes with a problem: what algorithm can complete such a task?

In this study, we attempt to solve this problem using a class of algorithms called Hierarchical Multilabel Classifiers(HMCs). As the name suggests, these are a kind of classifier that can assign multiple labels to one datapoint (repository), but the labels must be organized in a hierarchy. We present 4 different hierarchies and 4 different HMCs to see which combination yields the best results. These combinations are also compared to a non-hierarchical baseline. We find that HMCN-F, one of the HMCs, manages to marginally outperform the baseline with a difference in AUPRC scores of 0,024. While not a groundbreaking result, it is promising, as other methods of creating hierarchies may be able to beat the baseline by a larger margin.