Research on intelligent evaluation system of program algorithms used in IT development process

For traditional IT development work, besides project design and development team, software testing engineers are also needed for code testing and quality management, which confronts much redundant and high-cost works. With the rapid development of artificial intelligence, it is possible to find a new way for analyzing and evaluating source codes automatically and help to reduce redundant works and R&D costs caused by manual work. Based on the collection of source code data, this paper firstly describes the composition of structured programming corpus, and then analyzes code structure at the abstract syntax tree (AST) level in order to extract key features of those codes. Thirdly, a graph embedding-based vectorization algorithm is used to convert code and its abstract syntax trees into matrices and numerical representations. Fourthly, dimensionality reduction algorithms or optimization methods directly used are designed and applied for multidimensional process and principal feature extraction. Lastly, an adaptive hierarchical clustering method is used to achieve optimal partitioning of data samples. This paper elaborates basic principles, main components and implementation for constructing intelligent evaluation systems of source codes and the final experiment results verifies this intelligent evaluation system could not only learn significant differences between correct code and erroneous code, but also provide good clustering performance and surpass the typical unsupervised learning algorithms, such as K-means, DEC, DAEGC.