Research On Topic Identification And Evolution Of Customer Service Hotline In Telecommunication

Customer Service Hotline is an important channel for enterprises to obtain timely voice of the customers. For a long time, due to the limitations of technical means, data analyzing of the customer service hotline only focus on structured data such as traffic, customer satisfaction and so on, but there is a few work to find out potential value about the unstructured data which is the text from the voice. With the explosive growth of customer service hotline traffic, the expansion of the category and scope of user complaints, both identifying complaints quickly from the massive hotline text data and judging the user's complaint emotional evolution trend are important practical problems to be solved for customer service staff. The text mining of customer service hotline belongs to the research field of opinion mining, but the existing data mining is mainly based on Internet text data and the study of text message mining for customer service hotline is rare. The research of this paper is of great theoretical significance in extending the research scope of opinion mining and verifying the applicability of relevant theories and methods.Based on the theory and method of opinion mining, the language R is used as a programming tool to analyze semantic and emotion aspects of customer service hotline text information which comes from a telecommunication operator in China from September 2013 to September 2014,with the purpose to achieve automatic topic identification and prediction of topics combine with emotion. Specifically, at the semantic analysis level, more than 70 million text records are automatically classified as 20 topics by using Structural Topic Modeling (STM) algorithm. At the emotion analysis level, first of all, based on emotional word bank, and text emotion intensity algorithm, summarize the distribution of topic content and emotional tendencies of hotline text, then, get the trend of emotion tendency of the 20 topics by using time series autoregressive analysis method, and the feature of topic Emotional Evolution about different types of hotline text.Based on the work above, on the one hand, we construct the framework of the opinion mining for the customer service hotline context of the communication industry, on the other hand, we verify the structural topic modeling algorithm, the text emotion polarity algorithm and the autoregressive forecasting method based on the emotion polarity are all applicable for text semantic mining and emotional mining of customer hotline.At the practical level, the R program has realized the automatic identification and classification of the text topic of the customer service hotline, and the trend prediction of the emotional tendency of the text topic, so that expanded the new path of decision-making for the operators based on hotline text data.In the future, the research can be refined from following two aspects which contain both dimensional diversity and quasi-real-time analysis: on the one hand, considering adding other "metadata" such as complainant,complaint location and problem level to the topic model, enriching semantics mining diversity; on the other hand, combine the current single R program with other distributed systems such as Spark to enhance the quasi-real-time property.