Customer Churn Reduction Based on Action Rules and Collaboration

Abstract

Action rule mining is an important technology that can be applied to build recommender systems for reducing customer churn. Confidence, support, and coverage are used to measure the quality of action rules. In practice, action rules with higher confidence and support are more useful to users. However, there is little research work focused on improving the quality of the discovered action rules. To improve the quality of action rules extracted from a given client, this article proposes a guided (by threshold) agglomerative clustering algorithm by utilizing the knowledge extracted from semantically similar clients. The idea is to pick up only such clients that are doing better in business than the given client and are semantically similar with the given client. By doing that, the given client can follow business recommendations from the better-performing clients. The algorithm is guided by the threshold value checking how large the improvement of action rules discovered so far in their confidence is. If the improvement is lower than this threshold, the algorithm stops.