Customer churn leads to higher customer acquisition cost, lower volume of service consumption and reduced product purchase. Reducing the outflow of the customers by 5% can double the profit of a typical company. Therefore, it is of significant value for companies to reduce customer outflow. In this dissertation research, we mainly focus on identifying the customers with high chance of attrition and providing valid and trustworthy recommendations to reduce customer churn.
We designed and developed a customer attrition management system that can predict customer churn and yield actionable and measurable recommendations for the decision makers to reduce customer churn. Moreover, reviews from leaving customers reflect their unfulfilled needs, while reviews of active customers show their satisfactory experience. In order to extract the action knowledge from the unstructured customer review data, we thoroughly applied aspect-based sentiment analysis to transform the unstructured review text data into a structured table. Then, we utilized rough set theory, action rule mining and meta-action triggering mechanism on the structured table to generate effective recommendations for reducing customer churn. Lastly, in practical applications, an action rule is regarded as interesting only if its support and confidence exceed the predefined threshold values. If an action rule has a large support and high confidence, it indicates that this action can be applied to a large portion of customers with a high chance. However, there is little research focused on improving the confidence and coverage of action rules. Therefore, we proposed a guided semantic-aided agglomerative clustering algorithm to improve the discovered action rules.