Software bug prediction using graph neural networks and graph-based text representations

Ilias Siachos, Nikos Kanakaris, Nikos I. Karacapilidis

Abstract: Highlights•A novel approach for identifying and predicting software bugs utilizing graph attention networks.•Inductive learning on graphs, using Graph attention networks.•A robust method relying only on the structural and textual characteristics of a corpus of documents.•Fusing techniques from Graph Neural Networks and word embeddings to facilitate graph classification.•Dealing with the problem of text classification for short-text data.