Context-Enriched Multi-Intent Customer Dialogue Repository for AIDriven Assistance

Abstract

The global conversational Artificial Intelligence (AI) market is anticipated to reach USD 32 billion by 2030, with chatbots expected to manage over 80% of customer interactions. Despite this growth, manual annotation and classification of customer intents remain labor-intensive and inconsistent, limiting the scalability of customer support systems. To overcome these limitations, this study introduces an advanced Natural Language Processing (NLP) framework based on a Customer Support Bitext dataset labeled with multiple intents and categories. The approach begins with NLP preprocessing and Exploratory Data Analysis (EDA) to standardize, tokenize, and examine data distributions. Subsequently, Miniature Language Model (MiniLM) is utilized for efficient yet context-aware feature extraction. To address class imbalance within the dataset, the Synthetic Minority Over-sampling Technique (SMOTE) is applied to generate synthetic samples for underrepresented classes. In contrast to traditional models such as Decision Tree Classifier (DTC), K-Nearest Neighbors (KNN), and Naïve Bayes Classifier (NBC), the proposed framework combines Deep Neural Network (DNN)-based feature selection with KNN for improved classification performance. The system is designed to predict two bivariate outputs Intent and Category thereby enhancing contextual interpretation of customer queries. Finally, the model is deployed within a chatbot interface to enable real-time intent recognition and automated responses. The proposed framework improves chatbot accuracy, minimizes annotation inconsistencies, and enhances customer satisfaction through effective multi-intent understanding. This scalable and efficient solution highlights the effectiveness of integrating advanced embeddings, data balancing techniques, and feature selection methods to advance conversational AI in customer support systems.