Clone code detection method capable of automatically constructing and utilizing pseudo-clone corpus

Abstract

The invention discloses a clone code detection method capable of automatically constructing and utilizing a pseudo-clone corpus. The clone code detection method comprises the following steps: step 1,automatically constructing a pseudo-training corpus; step 2, determining the most effective code unit in the clone code detection task, namely, segmenting a code basic unit required by clone detectionby using a BPE method; step 3, splicing tokens according to the most basic and most effective code units BPE determined in the step 2, performing word embedding, training a word embedding model for all methods in a corpus by using a statement sequence displayed by token characters, and generating a {code unit-vector} corresponding dictionary; step 4, establishing a simple and effective BiLSTM classification model for code clone detection, and performing training; and classifying the two methods by utilizing an L2-Norm algorithm, and judging whether the two methods are clone pairs or not. According to the invention, a clone code detection tool is realized, and a better detection effect on 1, 2 and 3 type clones is achieved.

Description

technical field [0001] The invention relates to the fields of program analysis and machine learning, in particular to a program code clone detection method. Background technique [0002] The prior art involved in the present invention is as follows: [0003] (1) Clone code detection: Clone code refers to code fragments with similar text or similar functions in software projects. Studies have shown that the proportion of clone code in a system is about 7% to 23%, or even as high as 50%. According to the difference in code similarity, code cloning can be divided into four types: Type 1 cloning: After excluding the differences in the format of program codes such as newlines, blanks, and tabs, as well as the differences in comment statements, the two code fragments Exactly the same; type 2 clone: ​​two code fragments meet the definition of type 1 clone except for the difference between constant values ​​and identifiers such as variable names and function names, then these two c...

AI Technical Summary

Problems solved by technology

However, there is currently a lack of detection algorithms with high accuracy and good space-time efficiency, and the detection of code clones is still a relatively difficult research topic.

Images