Intelligent Query Understanding Using Intent Classification, Sentiment Analysis, and Automated Response Generation

Abstract

With the rapid growth of online services, the demand for intelligent, responsive, and emotionally aware customer service systems has increased significantly. This research proposes a hybrid natural language processing (NLP) framework that combines intent classification, sentiment analysis, and automated response generation for user queries in customer support domains. The proposed system leverages both traditional machine learning models (Logistic Regression, SVM, Random Forest, Naive Bayes) and modern transformer-based models (BERT for classification and T5 for response generation). The system classifies user input into distinct categories such as order issues, delivery concerns, technical support, and account management using TF-IDF-based traditional ML models, with BERT enhancing contextual understanding. Sentiment analysis using TextBlob assigns emotional polarity (positive, negative, or neutral) to each query. Rule-based and T5-generated responses are then dynamically adapted to reflect intent and sentiment.
Experimental results show that BERT+ T5 model achieved the highest accuracy (99.87%) among models, while SVM slightly outperformed it in complex queries. T5 provided more natural and personalized responses than templates. This integrated solution offers scalability, emotional intelligence, and accuracy, making it ideal for real-time support systems.