Student Program Classification Using Gated Graph Attention Neural Network

Source code mining has received increasing attention, among which classification plays a significant role in understanding and automatic coding. Most source efforts aim at the of projects, are usually large standardized, but less for student programs. There two differences between project codes On one hand, some work on is based relatively single information, far from enough Because programs small, makes them contain information. Consequently, it necessary to mine as much information possible other variable or function naming structure irregular, compared with projects. To learn programs, we proposed Graph Neural Network (GNN) model, integrates data flow call Abstract Syntax Tree (AST), applies an improved GNN model integrated graph achieve state-of-art program accuracy. The experiment results have shown that can classify accuracy over 97%.