Research On Telecommunication Complaint Texts Based On Text Classification And Emotional Score

With the increasingly expanding communications industry and increasingly fierce competition today,when users are dissatisfied with the services provided by operators,complaints will be generated.In order to improve user satisfaction and prevent user loss,operators need to efficiently and quickly respond to complaints and process it.According to the industry complaint handling process,proposes text classification to automatically classify complaint texts according to the responsible departments(such as channel-service,market,network,information security,government and enterprise,and support).For each category,obtain the emotional scores by the emotional dictionary,so that the operator staffs can effectively deal with the problem,avoiding the problem circulation and negative emotional users because the problem is not solved in time,resulting in the escalation of complaints or loss users.Firstly,observing the collected telecommunication complaint text sets,randomly select 10000 complaint texts for each category,extract the complaint text information and complaint category.Then,texts are segmented by dictionary segmentation and professional words in the field of telecommunication,word frequency statistics for each category,structure bag of words model and PLDA topic model,which are transformed into a classification problem with the topic as the independent variable and the complaint category as the dependent variable.Next,using the logistic regression,KNN,Naive Bayes and decision tree,as well as GBDT and random forest to model the training sets,predict it on the testing sets,evaluate model according to the index such as accuracy,recall,F1-measure and Kappa value.Finally,for different types of complaint texts,calculate the emotion scores based on sentiment dictionary.The construction of sentiment dictionary is composed of HowNet Emotion Dictionary and professional emotion dictionary on telecommunications field.According to the emotional calculate rules,get the complaint text emotion score.The findings of this study show that the classification based on the responsibility attribution department is a rule that automatically classify and effectively solve the complaint problem.Using the PLDA topic model can reduce dimension,and it is applicable to the classification problem of the complaint text from the result of logistic regression.And the accuracy of text classification constructed by random forest based on PLDA theme model is 81.61%,and the kappa consistency is 77.93%,it is a good predict method.In addition,the emotional score based on sentiment dictionary can judge the emotional level very well.Machine learning algorithms and emotion scoring rules can facilitate the submission and response of complaints.The classification algorithm and sentiment analysis in this paper can help operators deal with complaint classification and resolve complaints,improve user satisfaction,reduce loss,and protect consumer rights.