Research And Implementation Of Classification Method For Noise Work Tickets

Trouble ticket is the product of large and complex IT system,which records the event information generated by the system or reported by users,and is an important data carrier for system operation and maintenance activities.It is one of the important research contents of intelligent system operation and maintenance to automatically implement ticket problem classification,which has a significant contribution to improving the efficiency and reducing the cost of system operation and maintenance.The existing ticket classification method does not consider noisy label problem,where have noise in the label,and the accuracy of the classification method is not high enough.Therefore,it is necessary to study the classification method for noisy label to eliminate the negative impact of noise label on the classification of ticket,and then improve the classification accuracy.In this paper,we study the methods of noise label recognition and relabeling in ticket dataset.On this basis,the method of ticket automatic classification based on deep learning is studied,in order to improve the accuracy of ticket automatic classification.The specific research work is as follows:1.In order to reduce the influence of noise label in ticket data,a relabeling method based on the random noise classification model is proposed.This method uses classifiers such as logical regression and na?ve Bayes to identidy the noise label data in the training dataset,and then relabel the most reasonable and accurate label for the noise data based on the maximum prediction probability,so as to get the clean ticket dataset.Experiments on synthetic dataset,text dataset and real world ticket dataset show that this method can effectively reduce noise rate and improve classification accuracy.2.In order to improve the accuracy of automatic classification of ticket,we proposed a ticket classification method based on the bidirectional recurrent reural network(Bi RNN).This method uses word embedding technology to represent the ticket data in vector form,and use these vectors as the input of the learning model.The model use Bi RNN to learn the forward sequence and backward sequence of the ticket text,and uses hierarchical attention network at word and sentence level to learn the key words and key sentences in the ticket text.Based on the classification of base classifier,we use the hierarchical multi-label classification algorithm to consider the hierarchical information between the multi-labels of ticket.The experimental results show that this method can effectively extract the key information of the ticket text and achieve the accurate hierarchical multi-label classification.3.Based on the above research,we designed two core modules of ticket management system,ticket preprocessing module and ticket classification module,by using the method of noise relabel and the method of ticket classification based on bidirectional recurrent neural network.