Gated Multi-task Learning for QoS-aware Service Recommendation

Abstract In the ever-increasing diversity of cloud environment, recommendation web service-related applications based on service quality is a basic tasks that both providers and developers are very interested in. A key challenge how to get personalized accurate QoS values services from multi-source information users/services. addition, complex contexts multi-tasking attributes also make collaborative prediction difficult meet. Inspired by progress multi-task deep learning, this paper proposes new framework called gated expert network (MGEN), which gating mechanism learning. The MGEN consists three parts: expert-based context-source feature extraction, interaction learning fusion. First, multiple networks used capture latent representation in raw data services, including implicit features matrix explicit descriptions. process, several independent experts extract different dimensions. Second, output embedding transmitted each task layer through weights, global local high-order interactions. Third, interactive fused splicing, then fed back module for final recommendation. Results comparative experiment demonstrate proposed significantly outperforms state-of-the-art models recommendation, where various sequences effectively extracted. performs well independently modeling task.