QoE-Aware Network Slicing and Resource Allocation for Stable Mobile Online Learning

Abstract

Mobile online learning is challenged by the inherent conflict between dynamically evolving pedagogical demands and rigid resource provisioning infrastructures. Conventional quality of service (QoS)-driven resource allocation paradigms suffer from two critical limitations: 1) limited responsiveness to temporal-spatial fluctuations in service requirements, leading to unstable learning experiences, and 2) insufficient incorporation of students’ preferences, which diminishes continued learning willingness. To overcome this dilemma, we propose a closed-loop intelligent framework integrating dynamic network slicing and quality of experience (QoE)-aware resource allocation. First, we design a hybrid convolutional neural network long-short term memory (CNN-LSTM) architecture that synergistically captures spatial-temporal behavior patterns, improving student demand prediction accuracy for requirement-oriented network slicing. Second, we develop a Lyapunov-based optimization mechanism that ensures system stability within each network slice while considering student demand. Third, we incorporate student preference profiling across different network slices, enabling progressive convergence toward a hindsight-optimal resource allocation strategy. This framework features a feedback-driven allocation mechanism guided by queue stability and student preferences, dynamically adapting to user demand to further enhance QoE. Experimental results validate the framework’s efficacy. Compared to state-of-the-art baselines, our solution enhances student demand prediction accuracy, resource utilization, and prolongs average learning engagement duration.