Analytics-driven complaint prioritisation via deep learning and multicriteria decision-making

Abstract

Complaint analysis is an essential business analytics application because complaints have a strong influence on customer satisfaction (CSAT). However, the process of categorising and prioritising complaints manually can be extremely time consuming for large companies. In this paper, we propose a framework for automatic complaint labelling and prioritisation using text analytics and operational research techniques. The labelling step of the training set is performed using a simple weighting approach from the multiple-criteria decision-making (MCDM) literature, while transformer-based deep learning (DL) techniques are used for text classification. We define two priority classes, namely, urgent complaints and other claims, and develop a system for automatic complaint categorisation. Our experimental results show that excellent predictive performance can be achieved with state-of-the-art text classification models. In particular, BETO, a bidirectional encoder representations from transformers (BERT) model trained on a large Spanish corpus, reaches an accuracy (ACCU) and area under the curve (AUC) of 92.1% and 0.9785, respectively. This positive result translates into a successful complaint prioritisation scheme, which improves CSAT and reduces the churn rate.