Sentiment analysis for Arabic call center notes using machine learning techniques

Call centers handle thousands of incoming calls daily, encompassing a diverse array of categories including product inquiries, complaints, and more. Within these conversations, customers articulate their opinions and interests in the products and services offered. Effectively categorizing and analyzing these calls holds immense importance for organizations, offering a window into their strengths, weaknesses, and gauging customer satisfaction and needs. This paper introduces an innovative approach to extract customer sentiments through an advanced sentiment analysis technique. Leveraging two distinct yet synergistic algorithms—Support Vector Machine (SVM) and Neural Networks (NNs)—on the Kaggle machine-learning platform, our method discerns the polarity of each note, classifying them as positive, negative, or neutral. To enhance the quality of our analysis, we employed Natural Language Processing (NLP) and a range of preprocessing tools, including tokenization. The dataset comprises three thousand notes from various telecommunication companies, authored during real call center interactions. These notes form the basis of a specialized corpus, notable for being composed in the Jordanian dialect. Rigorous training and testing procedures were conducted using this corpus. The results are notable: our proposed algorithms displayed strong performance metrics. SVM yielded a commendable accuracy rate of 66%, while NNs excelled, boasting an impressive accuracy rate of 99.21%. These achievements are substantiated by comprehensive confusion matrices. In conclusion, our research provides a novel and robust framework for customer sentiment analysis in call centers, underpinned by the fusion of SVM and NNs. This technique promises valuable insights into customer feedback, facilitating informed decision-making for businesses seeking to enhance their services and products.