Research On Defect Prediction Model Of Software Feature Mining

The main research content of software defect prediction is to use the historical defect detection data in the software development process,use scientific methods to guide,and model the software module to achieve the purpose of predicting defects.According to the prediction results of the software defect prediction model,it can guide software developers to allocate test resources reasonably,improve software development efficiency,and thus reduce software test cycles and software development costs.In practical application scenarios,researchers have found that for software defect prediction models,different software defect modeling algorithms,such as machine learning algorithms,do not have significant performance gaps under the same feature indicators,but under different feature indicators The performance difference of the software defect prediction model is relatively significant.However,most of the software data feature collection is through manual statistics,and the statistical process of data features requires a lot of manpower and material resources.So the researchers shifted the research focus to the data features of automated learning software modules.As neural network technology has made tremendous breakthroughs in the field of natural language,it is thought that the software module code contains rich semantic and grammatical information,so some researchers have applied natural language processing related technologies to the field of software defect prediction.From the experimental results,the software defect prediction model based on natural language processing technology is superior to the defect prediction model based on statistical features in terms of performance,which proves that the neural network technology can effectively mine the rich semantic and grammatical information in the software modules.However,there are still some limitations in the previous software defect prediction research.When the researchers mine the feature information of the software system,they treat the software modules in the software system in isolation,or just consider the relevant statistical indicators of simple software modules during the inheritance process..However,from the perspective of the entire software system,a software network can be constructed according to the dependencies between software modules,etc.,and the data characteristics of the software modules can be analyzed and analyzed from the perspective of the software network.In order to solve the problem of limited extraction of structural features of software systems during the process of software system feature mining,a hybrid defect prediction model based on network representation learning was proposed in the study.The code semantic mining defect prediction model based on convolutional neural network can not only automatically mine the rich semantic information of software program modules,but also perform end-to-end software feature extraction.On this basis,the research goes further and analyzes the software network structure that the software system converts according to the dependency relationship.First,the software program system uses modules as the basic unit to generate the corresponding software network;then,the network characterization technology is used to automatically mine the hidden system structural features of each module in the software network;finally,a defect is constructed by integrating semantic features and system structural features Forecasting model.Through comparative test evaluation,the structural features of the software module system proposed by the research can effectively improve the performance of the software defect model.