ABSTRACT
Software defect prediction is used in the program maintenance process to find potential flaws in order to increase software reliability. A useful procedure that guarantees that the time and expense of software testing can be cut is the prediction of problematic software modules prior to testing. However, the majority of existing models either neglect a source codes tree structure or simply pay attention to a tiny portion of it; thus they do not fully exploit a source code. An abstract syntax tree (AST) of the source code for a software module contains information about the nodes and edges, and Graph Neural Networks (GNNs) evaluate this information to determine whether the module is defective or not.
Keywords: geometric deep learning, graph neural networks, reliability, software-engineering, abstract syntax tree
