CAMICS: A Context-Aware Multi-Intent Conversational System for Enhanced AI-Driven Customer Interaction Models

Customer interaction models are rapidly evolving with the advent of AI-driven chatbots and virtual assistants, yet challenges remain in achieving contextually aware, efficient, and scalable systems. This paper presents the Context-Aware Multi-Intent Conversational System (CAMICS), a novel architecture that integrates context encoding, multi-task learning, and efficient response generation to transform customer service interactions. Leveraging the Customer Support on Twitter dataset, CAMICS performs better state-of-the-art models across multiple tasks. For intent classification, CAMICS achieves an F1-score of 94.5%, surpassing BERT (91.0%) and CNN-based approaches (86.8%). In sentiment analysis, it obtains a macro F1-score of 92.1%, significantly outperforming logistic regression (77.5%) and Random Forest models (80.7%). The response generation module achieves BLEU and ROUGE-L scores of 56.7% and 54.9%, respectively, demonstrating its capability to generate high-quality, contextually relevant responses compared to GPT-2 (51.4% BLEU, 49.3% ROUGE-L). The CAMICS model also exhibits computational efficiency, with an inference time of 11.5 ms per conversation and reduced GPU memory usage (490 MB), making it suitable for real-time applications. An ablation study further highlights the critical contributions of the context encoder and multi-task learning modules to the overall performance.