Text mining of CRM data using decision tree induction and support vector machines.

• Main Author: Cheung, Chee Wai.
• Other Authors: Khoo Soo Guan, Christopher
• Format: Theses and Dissertations
• Language: English
• Published: 2010
• Subjects: DRNTU::Engineering::Computer science and engineering::Information systems::Information storage and retrieval
• Online Access: http://hdl.handle.net/10356/42237
• Institution: Nanyang Technological University

In today's business environment, the organization's call centre has become an important reservoir of customer intelligence. This intelligence can be tapped upon to develop better products and service approach to enhance customer satisfaction and improve the organization's bottom-line. This study seeks to develop an automatic method to categorize the call centre CRM data of a consumer electronics manufacturer. The study contributes to the existing literature by comparing and evaluating the text mining modeling techniques of Decision Tree Induction and Support Vector Machines as applied to call centre CRM data in free text form. In the sphere of feature selection, the impact of bigrams, extracted keywords and expert selected keywords are discussed. The Decision Tree Induction model produces prediction accuracy of 52.74% and the SVM model generated prediction accuracy of 60.4%. While the SVM model provides the higher prediction accuracy, the Decision Tree Induction model has demonstrated its usefulness in the interpretation and explanation of the results, especially in understanding the impact of influential keywords on specific target topics. The recommendation is to incorporate both the Decision Tree Induction and SVM models in the automated process of categorizing the call centre data. The inclusion of bigrams and expert selected keywords would have a positive impact towards the text mining process. The limitations of this study include the need to involve domain experts, overcome the spelling and grammatical errors in the dataset and the small number of cases for certain target topics.