A Deep Learning Method For Conditional Expression Semantic Error Location And Repair

With the increasing of computers applications,people are beginning to learn pro-gramming all over the world.When learning programming,people often use the online judgment system for programming exercises.In the process of writing a program,you will encounter a variety of errors and bugs that need to be debugged.These errors and bugs can usually be divided into syntax errors and semantic errors.People need to spend a lot of time debugging the code,which is very cumbersome and difficult for people who are just beginning to learn programming.In particular,semantic errors in conditional expressions are difficult to debug.However,the current online judgment system does not correct and guide the user's code.In recent years,machine learning and deep learning has made great sucess in many fields such as natural language processing,machine translation,speech recognition,and computer vision.This has also brought new methods for automatic modification and generation of program.Therefore,This thesis studies the semantic error correction and repair in the conditional expression of code based on the deep learning method.The main research work and innovations are as follows:1.We proposed a code semantic error location method based on the attention mech-anism deep neural network model.Our model combines the deep neural network model of the attention mechanism with the pointer network.After inputting the semantic error code of the program code into the model,the line number and position of the code where has semantic error in conditional expressions can be obtained.After relevant experimental experiments,it is verified that the model can better locate errors in conditional expressions.2.We proposed a method for inputting abstract syntax tree in the program as a fea-ture into the deep neural network model.In the preprocessing of the code data,we extract the hidden information contained in the code more completely.In addition to the code text in the program,we also add the abstract syntax tree information of the code.This information is not used in many program semantic error auto-repair methods.Not only did they ensure that the program code af-ter the repair was executable,but it also significantly improved the accuracy of automatic fixes for program semantic errors.