An Efficient Fully Connected Tensor Network for QoS Prediction

With the rapid development of Web services, quality of service (QoS) prediction has become a key technology to guarantee user experience and network resource optimization. In the past, many methods have been proposed for QoS prediction. Among them, models based on tensor decomposition are commonly used for QoS prediction based on time-varying historical QoS data. However, most tensor decomposition-based models are based on simple decomposition methods, which cannot understand the deep feature information well due to their limited feature space and weak connection strategy between tensor modes. Therefore, in this work, an efficient fully connected tensor network is proposed for QoS prediction. And fully connected tensor network is particularly good at characterization. However, fully connected tensor network is computationally complex. Considering that the target tensor is sparse, it can be solved using the single-element method. So we propose the methodology of this paper. Experiments on dynamic QoS datasets show that our model has higher prediction accuracy for missing data and is computationally efficient.