Clustering bank customer complaints on social media for analytical CRM via multi-objective particle swarm optimization

Abstract

The ease of access to social media and its wide reach has made it the preferred platform for consumers to express their opinions and grievances regarding all types of products and services. As such, it serves as a fertile ground for organizations to deploy analytical customer relationship management (ACRM) and generate business insights of tremendous value. Of particular importance is identifying and quickly resolving customer complaints. If not redressed, these complaints could lead to customer churn and when addressed quickly and efficiently, they can double the profits. This automatic product-wise clustering of complaints helps in better sentiment analysis on products and services. In this paper, two variants of a novel multi-objective clustering algorithm are proposed with applications to sentiment analysis, an important area of analytical CRM in banking industry. The first variant, MOPSO-CD-Kmeans, employs Multi-objective Particle Swarm Optimization along with heuristics of K-means and the second variant, MOPSO-CD-SKmeans employs the same multi-objective particle swarm optimization along with the heuristics of Spherical K-means to find an optimal partitioning of the data. Two clustering criteria were considered as objective functions to be optimized to find a set of Pareto optimal solutions, and then the Silhouette Index was employed to determine the optimal number of clusters. The algorithm is then tested on bank based complaint datasets related to four Indian banks. Experiments indicate that MOPSO-CD-SKmeans is able to achieve promising results in terms of product-wise clustering of complaints and could outperform the first variant.