A Study of Knowledge Extraction from Free Text Data in Customer Satisfaction Survey

Abstract

The progress of information technology has enabled companies to collect and process enormous amount of information. The direct information from customers, such as complaints and requests, is one of the valuable information for companies to increase the customer satisfaction and accordingly to retain the customers. Open-ended responses in questionnaire surveys contain important signals and suggestive ideas from customers. However, companies have not made full use of this type of information. This is because there is no decisive method to accurately extract the opinions and the demands of individual customers from vast amount of free text data. The analysis of free text data requires huge amount of time, effort, and expertise in the specific area. Moreover, there is a problem that the results of analyses may differ vastly between the analysts.

This paper proposes a method to extract potential unsatisfied customers by applying text mining to customer satisfaction analyses. The data we use for the analyses is a customer satisfaction survey conducted on corporate users of the Internet circuits. The survey consists of three parts: the satisfaction rating part (where the users scores the satisfaction levels of twelve factors in five scales), the reason description part (where the users describe the reasons of the ratings in free text format), and the overall comment part (free text format) at the end of questionnaire. We firstly construct a knowledge base of customer satisfaction factors using the satisfaction rating and reason description parts. Then we analyze the overall comment part using the knowledge base by 1) extracting the expressions of satisfaction factors, 2) classifying the expressions automatically based on the themes and contents, and 3) scoring the satisfaction level of the customers. As a result of evaluating the analyses results, we confirm that the method enables us to automatically classify the information about customer satisfaction in vast amount of free text data, which reduces the costs of analyses in time and in human resources. The result of evaluations also indicates the possibility of extracting potential unsatisfied customers automatically.