Feedback-Driven Coevolutionary Hybrid Intelligence for Personalized Recommendation in Smart Home Social Internet of Things: Enhancing User Satisfaction through Adaptive Learning

Abstract

The growing adoption of smart home devices within the Social Internet of Things (SIoT) has increased the need for adaptive recommendation systems that can respond to changing user preferences and device interactions. While existing studies primarily focus on recommendation accuracy and network modelling, limited attention has been given to user feedback-driven optimization for enhancing long-term user satisfaction. This study introduces the Coevolutionary Hybrid Intelligence Framework for Personalized Recommendation in Smart Home SIoT (CHI-PR-SIoT), incorporating a Human-in-the-Loop (HITL) mechanism that continuously refines recommendations using explicit user feedback. The framework combines a Graph Attention Network (GAT) for modelling social relationships among devices, a Bidirectional Gated Recurrent Unit (BiGRU) for capturing temporal context, and a Coevolutionary Particle Swarm Optimization with Adaptive Mutation (CPSO-AM) for feature and hyperparameter optimization. Experimental evaluation in a smart home SIoT environment involving Curtain, Light, Thermostat, Geyser, and Coffee Maker devices demonstrates that retraining with 250 user-labelled feedback samples improves prediction accuracy across all categories, achieving gains of up to 4.8%. Furthermore, the framework attains a user satisfaction score of 0.8456, outperforming the best baseline by 6.6%, highlighting the effectiveness of structured feedback integration for personalized smart home recommendations. These findings underscore the practical value of embedding structured feedback loops within coevolutionary optimization pipelines for real-world smart home deployments, offering a scalable pathway toward user-centric, adaptive home-automation systems that remain robust under sparse and cold-start conditions.