Research On Classification Method Of Kernel Bug Report Based On Deep Learning

Natural language processing plays an important role in the field of deep learning.With the research of deep learning in this field and the continuous progress of computer hardware equipment,the optimization of text classification algorithm is also being carried out continuously.At present,the algorithms used for natural language processing mainly include some classic deep learning models such as CNN,RNN and CLSTM.Although these models can deal with common text data well,some common deep learning models cannot do a good job for unstructured text data with high noise such as Bug Report.For a bug report with detailed information,how to determine the bug type of the report is a critical part of the effort to fix the bug.After getting the bug type,assigning it to the corresponding developer to solve such bug can greatly improve the efficiency of bug processing.How to automatically classify becomes the key to improve efficiency.A bug report mainly consists of title and description information,and classifying this information presents a major challenge.Since the content of the Bug Report is dominated by unstructured text,code segments and stack information,these information leads to the high noise of the Bug Report data.Among the commonly used deep learning algorithms,the algorithm used to process natural language fails to achieve good classification accuracy in the face of such text data.Therefore,this paper uses A new model,which uses A deep bidirectional cyclic neural network based on attention mechanism(DBRNN-A)to process Bug reports.The principle is that the grammatical and semantic features of sentences can be learned from a long sequence of words in an unsupervised form,and the model can remember and focus on the important text parts in the bug report through the mechanism of attention.The algorithm can effectively deal with the context of the long word sequence,and can effectively deal with the synonym direct relationship,and better classification of the data set.In this paper,the main data set used is from the Bug Report provided by Syzbot.Syzbot is a fuzzy test report system based on Syzkaller Fuzzer for Linux kernel,and Syzkaller is a fuzzy tester guided by coverage rate.By processing the bug report provided by Syzbot,we obtained a benchmark data set and carried out corresponding experiments through this data set.In the experimental step,different algorithm models are used to carry out experiments on the same data set,and the performance of DBRNN-A model for automatic bug report classification is verified by setting experimental accuracy and differentiated data sets.The effect of input richness on model accuracy was tested by setting the combination of title attribute and description attribute in data set.