Research And Implementation Of Quality And Efficiency Evaluation Method For Automatic Code Generation Based On TBCNN

Automatic code generation can simplify the programmer's development work to a large extent,and can improve the efficiency of software development.Automatic code generation has also become a hot spot in current research.At present,part of the research on automatic code generation technology has been applied to actual development.Automatic code generation tools implemented according to a certain code automatic generation method are usually embedded in the integrated development environment in the form of plug-ins to help programmers improve development efficiency.The quality and efficiency of automatic code generation will directly affect the development efficiency of programmers.A perfect automatic code generation quality and efficiency evaluation can not only effectively evaluate the generated code,but also help to find the problems of the automatic code generation tool itself.Existing research lacks a unified evaluation method for the quality of automatic code generation.Most studies use Precision,Recall,MRR,and F1-Measure as indicators to evaluate the quality and efficiency of automatic code generation,but these indicators only consider the number of automatically generated codes and the correctness of the generated codes on the quality and efficiency.In the actual development using automatic code generation tools,there are many factors that affect quality and efficiency,and the behavior of the programmer also plays a major role in them.Existing evaluation methods do not take into account the dynamic change of the code context during the automatic code generation process,and only focus on limited attributes.Most studies use different evaluation indicators,and there is no direct conversion between the indicators,lacking a unified evaluation method,and it is difficult to compare various code automatic generation models and methods.In response to the above problems,this article mainly carried out the following research work:(1)In order to evaluate the quality and efficiency of automatic code generation,this article analyzes the process and results of automatic code generation,and establishes the quality and efficiency evaluation of automatic code generation by combining programmer behavior in the process of automatic code generation and the characteristics of automatic code generation tools.model.(2)Through the automatic code generation process,analyze the programmer's behavior and the characteristics of the automatic code generation tool information that may affect the quality and efficiency of automatic code generation.Based on the above characteristics,a method for monitoring programmer behavior and automatic code generation tool information is proposed.Considering the influence of the code context on the quality and efficiency of the generated code,this paper defines a MAST(Multidimensional Abstract Syntax Tree,MAST)feature tree,which uses the information in the MAST to express the features that exist in the automatic code generation process.Use TBCNN(Tree-Based Convolutional Neural Network,TBCNN)to extract features from MAST,and map the extracted features to the code using a fully connected neural network to automatically generate quality attributes and efficiency attributes.By assigning weights to the quality and efficiency attributes of the automatic code generation,the quality and efficiency evaluation of the automatic code generation is realized.(3)A data acquisition tool for programmer behavior and code automatic generation is realized.Based on the process of automatic code generation quality and efficiency evaluation,a prototype tool for automatic code generation quality and efficiency evaluation is realized,and the validity and rationality of the method of automatic code generation quality and efficiency evaluation proposed in this paper are verified through experiments.