DISTINGUISHING AUTHENTIC AND INAUTHENTIC EMPATHY IN CUSTOMER SUPPORT CONVERSATIONS

Abstract

Empathy plays a vital role in customer support and service experiences, as it helps companies connect with customers on a deeper emotional level. However, not all empathetic responses are perceived as genuine — some are authentic, while others are superficial or inauthentic. The primary objective of this research is to classify emotional expressions present in customer dialogues and use this classification to understand and analyze authentic versus inauthentic empathy in customer interactions. Using the Empathetic Dialogues dataset, we conducted a comprehensive study by implementing and comparing multiple machine learning algorithms to classify emotions expressed in textual conversations. We trained and evaluated several models, including Random Forest, LinearSVC (Linear Regression using SVM), and logistic regression, to classify emotions in the context of empathy-based customer segmentation. Among these, Random Forest performed the best, achieving an accuracy of 88.37%. This result highlights the model’s ability to effectively capture emotional patterns from textual data, making it a suitable choice for detecting authentic and inauthentic expressions of empathy in customer interactions.