Graph neural networks improve code-to-architecture mapping

Researchers have developed a semi-supervised node classification method to map code to architectural modules, addressing architectural technical debt. This approach utilizes a directed multi-relational dependency graph, with files as nodes and dependencies as edges. Experiments show that graph convolutional networks (GCN) and graph attention networks (GAT) significantly outperform a multi-layer perceptron (MLP) and a naive Bayes classifier on large codebases like Google Chrome, with GAT achieving the highest F1 scores. AI

IMPACT This research offers improved methods for understanding and managing software architecture, potentially aiding developers in large-scale projects.

RANK_REASON Academic paper detailing a new method for code-to-architecture mapping using graph neural networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Google Chrome
• Gotit.pub
• graph attention network
• graph convolutional network
• Hugging Face
• multilayer perceptron
• naive Bayes classifier
• ScienceCast

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →