Source code comparison method and system oriented to technical features and program product

Abstract

The invention discloses a technical feature-oriented source code comparison method and system and a program product, and belongs to the technical field of natural language processing. The method comprises the steps that a semantic coding method based on a function call structure is used, and code similarity is analyzed from the aspects of the function call structure, a function name, a built-in class and the like; carrying out graph semantic encoding by adopting an auto-encoder method based on a graph convolutional neural network, and comparing code structure semantics based on semantic vectors; based on the calling information vectors of the functions and the built-in class names, comparing the similarity of the functions with the similarity of the built-in classes; and finally, splicing the structure vector, the function vector and the built-in class vector as an overall technical feature vector, and comparing the code similarity. According to the method, technical feature information such as function names, calling structures and built-in classes is comprehensively considered, and code comparison can be better carried out according to technical features.

Description

Technical field [0001] The present invention discloses a technical code -oriented source code comparison method, system and program products, which belongs to the technical field of natural language processing. Background technique [0002] The open source platform provides the environment of sharing and communication code for scientific researchers. More and more deep learning models and code sharing on the open source platform, creating a reused code ecological environment. Therefore, for specific problems, researchers need to find out to find out Related solutions. Modern algorithm design ideology is a modular construction code, which usually contains a large amount of basic function functions. Its function names, call structure, and built -in category information provides important code technical features. For a variety of solutions for the same problem, such as in text classification, different neural network structures can be used: convolutional neural networks, circulating...

AI Technical Summary

Problems solved by technology

[0004] The code similarity calculation method based on the call structure is mainly to calculate the similarity between codes based on the logical structure of the code. The logical structure generally includes an abstract syntax tree and a program call graph. The source code similarity evaluation method based on semantic information at functional granularity, by calculating the identifier corresponding to the function and the embedding vector of the control flow graph, this method considers the structure and node information of the control flow graph, but this method does not consider the code import. Semantic information for technical features such as built-in classes

[0005] The code similarity calculation method based on static features is to extract some measurement values ​​from the source code to form feature vectors, and then use the similarity between the code feature vectors as the code similarity, for example: the invention patent with the publication number CN111290784A discloses the applicable Based on the program source code similarity detection method for large-scale samples, the local sensitive hash value is calculated for the text feature sequence and feature weight sequence of each sample to be tested, and this value is used as the sample feature vector. This type of method is generally applicable to most programs. code, but such methods generally do not consider the call information and structural semantics of the technical characteristics of the code

[0006] The binary-based code similarity calculation method generally disassembles the binary code to obtain the instruction sequence of each function, then vectorizes the instruction features, and finally calculates the code similarity through the feature vector, for example: the public number is CN113554101A The Chinese patent discloses a binary code similarity detection method based on deep learning, uses Structure2Vec to generate the graph embedding of the control flow graph of the binary function, and introduces CNN to process the sequence structure information between the basic blocks of the control flow graph, so as to better Clarify the sequence relationship between the internal blocks of the function. This method can well adapt to cross-architecture and cross-version similarity detection, but it does not consider the semantic information of technical features such as the function name of the code.

Images