Understanding Customer Satisfaction Levels in Call Center Conversations Through Data Programming

Abstract

In the telecommunications sector, customer satisfaction plays a crucial role due to the industry's exceptionally high churn rates. However, measuring satisfaction directly from customer feedback can lead to inconsistency and bias. This feedback tends to be polarized, either very positive or very negative. Moreover, as subjective evaluations, they can vary significantly from person to person. Consequently, it is challenging to establish a meaningful relationship between customer satisfaction survey scores and actual feeling. To obtain a more accurate assessment of customer satisfaction, we analyzed the dialogue between customers and call center representatives, which yields more reliable results. We investigated multiple factors influencing customer satisfaction, complementing the traditional customer rating approach. Specifically, we employed data programming techniques to develop a more robust customer satisfaction prediction model for Turkish Language. This methodology enabled us to train a NLP model, resulting in a more precise representation of customer satisfaction with weak supervision. This paper outlines the design and implementation of this innovative process.