Sentiment Analysis of Incoming Calls for Helpdesk

In the evolving landscape of customer service, understanding the emotional tone of interactions is vital for delivering empathetic, efficient, and personalized support. This thesis proposes a real-time sentiment analysis system for incoming helpdesk calls, transforming raw audio into actionable emotional insights. The framework integrates Automatic Speech Recognition (ASR) to transcribe calls, followed by Natural Language Processing (NLP) and machine learning-based sentiment analysis to classify emotions as positive, negative, or neutral. The system architecture features a Python- based backend pipeline, a Flutter-based customer feedback interface, and a JavaScript- driven admin dashboard for real-time monitoring. Hybrid modeling approaches—combining lexicon-based methods, traditional machine learning algorithms, and deep learning models like Transformers—are used to capture both linguistic and acoustic sentiment features. The system facilitates early detection of customer dissatisfaction, optimizes call handling, and identifies recurring issues and sentiment trends. Performance evaluation with synthetic data demonstrates its reliability and responsiveness. Key challenges addressed include transcription accuracy, emotional nuance detection, and ethical concerns related to privacy and bias. Future enhancements aim to support multilingual sentiment analysis, CRM integration, and finer-grained emotion detection, offering a scalable and ethically sound framework for embedding emotional intelligence into helpdesk operations.