摘要：With the advent of Big Code, code prediction has received widespread attention. However, the state-of-the-art code prediction techniques are inadequate in terms of accuracy, interpretability and efficiency. Therefore, in this paper, we propose a graph embedding model that integrates code semantic features. The model extracts the structural paths between the nodes in source code file’s Abstract Syntax Tree(AST). Then, we convert paths into training graph and extracted interdependent semantic structural features from the context of AST. Semantic structure features can filter predicted candidate values and effectively solve the problem of Out-of- Word(OoV). The graph embedding model converts the structural features of nodes into vectors which facilitates quantitative calculations. Finally, the vector similarity of the nodes is used to complete the prediction tasks of TYPE and VALUE. Experimental results show that compared with the existing state-of-the-art method, our method has higher prediction accuracy and less time consumption.
