Understanding Urban Spatio-Temporal Usage Patterns Using Matrix Tensor Factorization

Automated understanding of spatio-temporal usage patterns of real-world applications are significant in urban planning. With the capability of smartphones collecting various information using inbuilt sensors, the smart city data is enriched with multiple contexts. Whilst tensor factorization has been successfully used to capture latent factors (patterns) exhibited by the real-world datasets, the multifaceted nature of smart city data needs an improved modeling to utilize multiple contexts in sparse condition. Thus, in our ongoing research, we aim to model this data with a novel Context-Aware Nonnegative Coupled Sparse Matrix Tensor (CAN-CSMT) framework which imposes sparsity constraint to learn the true factors in sparse data. We also aim to develop a fast and efficient factorization algorithm to deal with the scalability problem persistent in the state-of-the-art factorization algorithms.