Formulation of an AI-Based Call Analytics Model for Analysing Mixed-Language Customer Calls

In this modern age, customer service management is often considered to require more digital communications, such as email and the web. Phone calls are the most reliable way for customers to request services and information directly at an instantaneous rate. However, for companies operating in Malaysia, most of the phone calls received are in Bahasa Malaysia mixed with English (we call this “Manglish”), whereas the existing software focuses on a single language. This research proposes a model to transform the audio of customer calls into useful information such as topics, complaints, service requests, inquiries, sentiments (positive, negative, neutral), and emotions. This research focuses on data collected from the electricity supply industry. The sentiment analysis experiment, conducted using a deep learning model, generates an accuracy of 92.86% for the Malay language and 75% for English. The results of the experiment reveal a word error count of 37.18% for the Malay language, 45.68% for English, and 59.65% for Manglish. For the topic classification experiment, deep learning (Neural Network), achieved 45.45% accuracy for Malay and 55.56% accuracy for English. An emotion recognition experiment recorded an accuracy of 89.58%. Some improvements to the existing model are also listed in this study.