MVLSC Home · Issue Contents · Forthcoming Papers

An Integrated Model Based on The Fuzzy Set Theory and Machine Learning Approaches to Evaluate Risk Levels of Customer Interview with A Real-Case Application in White Goods Industry
Gizem Köse And İhsan Kaya

Based on increasing in digitalization and technological advancements, the timely and efficient services are becoming increasingly important in the after-sales process to confirm customers’ expectations. It is essential to understand and respond to customers’ needs accurately and timely based on related analysis. To effectively analyze to these expectations and requests, it is beneficial to recognize their urgency, appreciate the emotions involved, and establish measurable criteria for the quality of communication with customers. In this paper, a model that can significantly enhance customer relationships and overall satisfaction has been developed that thoughtfully combines the Bidirectional Encoder Representations from Transformers (BERT) algorithm, a prominent machine learning technique, with the fuzzy theory (FST). The main aim of this study is to develop a model that concentrate more systematic and comprehensive approaches that take into account a number of factors, with the ultimate aim of improving customer satisfaction by reducing response times. This innovative approach enhances our ability to assess the quality of conversations conducted by customer representatives by utilizing Pythagorean Fuzzy Sets (PFSs) that allows us to achieve a more sensitive and measurable evaluation by using linguistic expressions. The BERT algorithm has been employed to effectively evaluate the criticality of the customers’ emotional state and service requirements. The proposed approach is demonstrated with accuracy rate of 72.7% with the BERT algorithm, surpassing the performances of Machine Learning Algorithms (MLAs) in processing text-based contents such as Naive Bayes that is 63.6%, Random Forest that is 54.5%, and Support Vector with is 66.6%. In the final step, a Fuzzy Inference System (FIS) has been constructed in the MATLAB environment to analyze the quality of the customer representatives’ interview, customers’ emotional states, and urgency of the service needed. This allowed us to thoughtfully categorize the risk level of customers’ conversations as “High”, “Medium”, or “Low” linguistic values. The suggested integrated model has been successfully applied on transcribed content of customers’ interactions at the call center within the white goods industry. The conversations identified as high risk by the model are directed to a specialized team, enabling a fast action to be taken. This proactive approach aims to enhance customers’ satisfaction level by addressing potential issues to avoid more problems. Furthermore, calls that exceed a certain threshold or are linked to significant negative feedback are forwarded to the complaint management team for appropriate resolution.

Keywords: Fuzzy Sets, Pythagorean Fuzzy, BERT Algorithm, Sentiment Analysis, Machine Learning, Fuzzy Inference System

Full Text (IP)