Telecom Customer Experience Analysis Using Sentiment Analysis and Natural Language Processing—Comparative Study

Abstract

In today’s competitive telecom landscape, service providers are increasingly seeking real-time customer experience analysis, prompt responses to customer feedback, and the ability to effectively promote new services. To achieve these goals, telecom operators are embracing digitalization initiatives that encompass the entire customer journey, broadly divided into three phases: engaging, using, and evaluating. Recent advancements in natural language processing (NLP) and sentiment analysis (SA) techniques have empowered telecom service providers to rapidly analyze and categorize millions of customer tweets, gaining valuable insights into service perceptions and user satisfaction. With a significant presence of telecom service providers in Arab countries, where customers frequently share their service experiences through Arabic tweets, the need for specialized NLP and SA techniques that can effectively handle Arabic language data becomes paramount. This study focuses on Arabic language processing and sentiment analysis to support one of the Middle East’s largest telecom service providers in analyzing and enhancing customer experience. The study successfully applied BERTopic, a topic modeling technique, to Arabic telecom-related text, generating six distinct clusters for 50% of the analyzed tweets. Support vector machine (SVM) outperformed XGBoost as the machine learning classifier when combined with the BERT-base model, achieving an F1-score of 0.71 compared to XGBoost’s 0.65. The fine-tuned MARBERT model demonstrated superior performance in text classification compared to machine learning algorithms, achieving an F1-score of 0.8571, a 3% improvement over the best-performing machine learning classifier.