Research And Application Of Duplicate Bug Reports Detection Using RNN

Research shows that dealing with a large number of repetitive bugs will cause great waste of manpower,especially for large projects.In order to reduce the workload of manual detection and repetitive report,and reduce the time of manual bug management,it is valuable to carry out the research on automatic repetitive detection method of bug reportThis paper systematically analyzes the causes,development status and significance of repetitive bug report,and summarizes the research methods of repetitive bug at home and abroad and the methods of deep learning in text similarity.Finally,based on the theory and technology of recurrent neural network and attention mechanism,the LSTM model,BI-LSTM model,BI-LSTM model and BI-LSTM model using attention mechanism are applied in the field of repetitive bug reporting to calculate the similarity between two bug reports.The bug reports produced by eclipse,an open source project,are used as data sets,and the data sources are selected by comparing with the previous experience to conduct experiments on the above models.After that,the results of each model are compared and analyzed,and some degree of optimization is carried out.The overall idea is to divide the bug reports into two categories:similar and not similar.When the calculated results of the two bug reports are greater than a certain threshold value,it is determined that the two texts are similar.In each model,different parameters such as learning speed,training rounds,data length are adjusted to get the best results.Finally,the double-layer BI-LSTM model based on attention mechanism can achieve the highest accuracy of 91.21%in this experiment.For the rest,the highest accuracy of LSTM model is 87.42%,BI-LSTM model is 88.94%,and double-layer BI-LSTM model is 90.75%In order to effectively reflect the function of duplicate bug report detection model,this paper implements a bug management system with duplicate detection as the core function.This system uses SSM framework and MySQL database to realize the basic functions of bug management,and applies the model to the bug management system in the way of timed tasks.Finally,five bug reports that are most similar to the bug report will be displayed on the bug report details page.