Design and Optimization of an Intelligent Mobile Interaction System for Adult Education

Abstract

Mobile learning scenarios in adult education impose mutually constraining technical requirements regarding real‑time interaction performance, recommendation accuracy, and data privacy protection. Neither cloud‑centric centralized inference nor device‑only local com‑ putation can adequately satisfy these demands. In this study, a terminal‑edge‑cloud collaborative intelligent mobile interaction system was proposed. A fuzzy‑inference‑based context‑awareness engine was deployed on the terminal device, performing local fused inference on temporal, spatial, and device‑status information to dynamically generate micro‑interaction strategies, with inference latency kept below 5 ms. The recommendation model was compressed to 20 MB through knowledge distillation and 8‑bit integer (INT8) quantization, enabling a single inference to be completed within 50 ms. At the edge nodes, user short‑term interests were maintained via differentially private gradients, while at the cloud side, model parameters were aggregated through federated averaging to achieve global evolution without raw data leaving the devices. This study provides a feasible technical pathway for the design of mobile interaction systems in resource‑constrained and privacy‑sensitive scenarios.