Dynamic QoS Prediction for Edge Intelligence with Multi-Modal Data Fusion

Edge intelligence services encompass a wide range of offerings that often perform similar functions, making Quality of Service (QoS) a crucial factor for selecting and recommending the best services to users. QoS metrics can fluctuate over time due to varying network conditions and service demands, making dynamic QoS prediction essential for maintaining optimal user experiences. However, current dynamic QoS prediction methods are limited as they either extend static techniques without fully capturing temporal changes and considering contextual information, or rely on time series methods that fail to effectively represent hidden features. To overcome these challenges, we propose a novel multi-view hybrid method for dynamic QoS prediction in edge intelligence services. This method integrates multi-modal data by capturing time-based dynamics with temporal dilated convolution, modeling contextual relationships with heterogeneous graph convolution, and uncovering latent feature interactions through temporal matrix factorization. Experimental results demonstrate that our method consistently outperforms several baseline techniques across multiple evaluation metrics, highlighting its effectiveness in dynamic edge intelligence environments.