Actionable Decision Support Model Based on Customer Tweet by Analyzing Emotion and Complaint from Bank XYZ

Abstract

Bank XYZ has more than 4.539 million mobile banking users in 2020, it's increased 40% YoY. By the time, bank XYZ focused on digital transformation by launching a super-app platform to encourage customers for moving to the online platform. To support that, social media Twitter is used to interact with the customer. If the tweet does not handle properly, it affects bank services. Multiclass Support vector machine (SVM) is used to extract customer tweets that contain complaints and emotions. The developed model provides an actionable decision that can decide an action to do based on customer tweet classification. The classification and constructed model is applied successfully to 2406 tweets from customers. As the result, by optimizing the soft margin (C) value from 0.1 to 1 by 0.01 stepping, the SVM classifier reaches the best accuracy of 77.36% on emotion with data validation for C value 0.62, and 72.54 on the complaint with cross-fold validation for C value 0.94. And actionable DSM provides action to 0.54% direct message and 99.46% reply tweet. Modification of SVM by optimizing the soft margin performs well with the dataset. Action can be suggested with rule-based actionable DSM to address customer complaints.