Research On Text Preprocessing And Summarization Technology For Customer Complaints In Telecommunication Industry

The rapid development of the communication industry has brought about the vigorous growth of telecom businesses and the continuous expansion of customer groups.While communication operators continue to pursue their idea of service first and customer foremost,the customer service center,one of its core business lines,is facing the test between the massive customer complaints and the traditional outdated handling methods.According to the feedback of a provincial operator that has been investigated,its current handling of customer complaints has shortcomings such as heavy labor,high cost,and low efficiency.Under the background of the continuous emergence of new services and the increasing number of complaints,disadvantages like rigid classification rules and subjective analysis are also showed.Thus,the research and application of text mining related technologies for in the telecommunications industry is of great significance on improving this situation.The topic of this thesis came from a cooperation project between school and China Telecom System Integration Co.,Ltd.In view of the current situation that text mining related researches are relatively sparse in the field of analyzing telecom customer complaints and the general technical solutions are not working effectively,the following work has been carried out in the thesis:1)Aiming at the problem that traditional Chinese text preprocessing technology is difficult to effectively extract and represent the text features of telecom complaints,this thesis proposed a text preprocessing optimization method for telecom complaints.The method involved three sub-steps for telecom complaints preprocessing as text cleaning,word segmentation and text representation.After cleaning the telecom complaints corpus,in the text segmentation step,this thesis realized the segmentation with an improving industry dictionary to target at the recognition of rich terminologies in telecom complaint texts;in the text representation step,considering the colloquial nature of the complaint content,this thesis optimized the influence of synonyms and stop words to reduce redundancy caused by invalid words.The experimental results have shown that,compared with the traditional method,the preprocessing method proposed in this thesis achieved a significant improvement in the quality of text preprocessing for telecom complaints.2)Aiming at the problem that the text summarization accuracy is not high due to the long length and unit structure of each clustered category of text,a combined long text summarization method based on length compression is proposed.Considering the fact that the current telecommunication business is complicated and lacks prior knowledge of the classification rules,which cannot use the text classification technology that was highly relied on in existing researches,this thesis first analyzed the clustering technology that is suitable for the content of complaints based on experimental verification.At the same time,to assist manual work on understanding the main content of each clustered category,this thesis built a self-labeled telecommunications industry customer complaint data set,and proposed a complementary text summarization method combining supervised generative and unsupervised extractive ways.The experimental results have shown that,when summarizing the clustered content of telecom complaints,the proposed method achieved better results than existing ones.