An MTVC–IOOA Framework For Semantic Feature Optimization In Intelligent Customer Support Chatbots

Abstract

Customer-support chatbots are now expected to understand diverse service requests and route users to suitable answers, and maintain quick response behavior under large volumes of queries. However, high-dimensional text representations often contain redundant and noisy features. This increases computational effort without necessarily improving intent recognition. This paper proposes a focused feature-optimization framework for intelligent customer-support chatbots by integrating Multiscale Transformer Vector Conversion (MTVC) with an Improved Osprey Optimization Algorithm (IOOA). The proposed IOOA uses Fuch chaotic mapping. The experimental study uses the Bitext customer-support dataset containing 26,872 samples and 27 intent categories. A representative 5,000-sample stratified experimental subset was used for optimization analysis. MTVC generated a 256-dimensional dense feature space, and feature-selection methods were evaluated at a 128-feature operating point. The proposed IOOA achieved 99.30% accuracy, 99.32% precision, 99.30% recall, and 99.30% F1-score while reducing the feature dimensionality by 50%. Compared with standard OOA, the proposed method improved accuracy from 98.40% to 99.30% and reduced execution time from 25.09 s to 13.88 s in the final evaluation. The result shows that chaotic enhancement can improve feature-selection quality and deployment efficiency when applied to structured customer-support intent classification.