Spatial-Temporal Aware Intelligent Service Recommendation Method Based on Distributed Tensor factorization for Big Data Applications

With the dramatic growth of public cloud offerings and heterogeneous data information, how to discover potentially valuable information from big history behavior data and design intelligent recommendation techniques has become more and more important. Due to the dynamics of cloud environment, both user behaviors and QoS (Quality of Service) performance of cloud services are sensitive to contextual information, such as time and location. However, the consideration of time and location information brings the increase in the order of rating matrix and the data sparsity problem. In view of these challenges, we propose a spatial-temporal aware intelligent service recommendation method based on distributed tensor factorization to address the above problems. First, the time and location information are introduced into the recommendation models by distinguishing time-sensitive QoS metrics and region-sensitive QoS metrics from stable QoS metrics. To deal with the sparse rating data, time slots and regions are clustered respectively. Then, a high-order tensor factorization technique is applied to mine the latent factors among users, services, time information, and location information. Moreover, to improve the scalability of our recommendation models in big data environment, a fast distributed asynchronous SGD (Stochastic Gradient Descent) mechanism is employed to get a good balance between the convergence speed and prediction accuracy. Finally, experiments based on both real-world data set and big synthetic data set are conducted to validate the effectiveness and scalability of our proposal. The experimental results show that our proposal achieves a good balance between the recommendation accuracy and scalability.