Contextual Preference Cognition through XLNet Semantic Encoding and Trivariate Decision Intelligence

Abstract

Customer support platforms generate an enormous number of service requests every day, making it difficult for organisations to analyse customer feedback and deliver timely responses. Extracting useful information from these text-based interactions is essential for improving service quality, customer retention, and operational efficiency. This research presents a multi-target prediction model that simultaneously determines the priority of a support ticket, classifies the expected customer satisfaction level, and predicts the possibility of successful issue resolution. Unlike conventional approaches that address each prediction task independently, the proposed framework exploits the relationship among these outputs to produce more reliable and consistent results. Existing customer support systems often depend on manual inspection, keyword-based techniques, or simple text analysis methods, which struggle to process large volumes of unstructured textual data and frequently produce inconsistent outcomes. To overcome these limitations, this work utilises the XLNet transformer model to generate rich contextual representations of customer messages. These features are evaluated using several classification algorithms, including Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), Stochastic Gradient Descent (SGD), and Nearest Centroid (NC). Their performance is compared with the Histogram-Based Gradient Boosting (HGB) classifier, which demonstrates the highest prediction accuracy and overall effectiveness. The proposed framework provides a scalable and intelligent solution for customer support analytics by enabling faster ticket assessment, more accurate satisfaction estimation, and improved resolution forecasting, thereby assisting organisations in making better service decisions and enhancing the overall customer experience.