An intelligent analysis method for software programs

Abstract

The present invention proposes an intelligent analysis method for a software program. The intelligent analysis method is an AST-based serialized neural network, and the AST-based serialized neural network is a hierarchical vector representation method. The implementation steps are as follows: Step 1-1: At the bottom layer, after converting the original program into AST, it is further divided to obtain the corresponding code block sequence. The code blocks are ordered and match the original statement order; Step 1- 2: Each code block uses a recursive encoder to encode this subtree into a vector e 1 , e 2 ,...,e t ; Step 1-3: Pass the obtained vector sequence through a bidirectional recurrent layer to extract the dependent features between code blocks; Step 1-4: All the multi-dimensional features of the time step are sampled by the pooling layer to obtain the final single vector representation.

Description

technical field [0001] The invention relates to an analysis method, in particular to an intelligent analysis method for software programs. Background technique [0002] Program comprehension plays a vital role in many software engineering applications. On the one hand, it can help developers improve development efficiency and code quality, and on the other hand, it can also help software maintenance. Typical applications include code classification, code clone detection, defect prediction, and code summary generation. Among them, the most important and difficult part of realizing program understanding is how to establish an effective program representation method to capture the semantics of the program itself as much as possible, such as the function of a program. [0003] Although we can manually design or extract program features to achieve many related applications, the disadvantage of this method is that it requires a lot of labor and the features obtained are often li...

AI Technical Summary

Problems solved by technology

[0005] 1. The neural network that only relies on the tree structure is easy to cause the problem of gradient disappearance;

[0006] 2. The method of traversing the entire AST from the bottom up or using a sliding window is easy to lose the dependency information between the code blocks contained in the original program;

[0007] 3. Most of the existing tree-based methods convert the original AST into a binary tree form, which destroys the original structural information, and will greatly increase the depth of the tree and further bring difficulties in capturing dependencies

Images