Classifying Packages for Building Linux Distributions
Qi Jing, Lian Yu, Lei Zhang, Jie Meng, Zheng Liu, Haolin Xue · 2023
Linux distribution provides users an open source operating system to fulfil a variety of purposes and missions. Currently, there is no package categorization to lead Linux distribution developers and manufacturers to select the components, and no architecture information to guide the construction process. This paper proposes to automatically sort out the packages and outline the boundaries for the component selection in terms of package categories, and provide a guidance for the architecture design in the construction. A package data set is created, and presented as a graph with the packages as nodes and package dependencies as edges. The realistic data set has the unbalanced or long-tail problem, and some labels tend to be similar with noises due to unavoidable reasons. This paper explores different models and approaches, adapts and optimizes the label representation techniques, introduces the contrastive learning mechanism, and improves the direct GraphSage model. The metrics of prediction started with less than 50%, and reaches more than 90%. This demonstrates the validity of the proposed approach. The data set is publicly available on Internet.