Time-aware tensor decomposition for sparse tensors

Given a sparse time-evolving tensor, how can we effectively factorize it to accurately discover latent patterns? Tensor decomposition has been extensively utilized for analyzing various multi-dimensional real-world data. However, existing tensor models have disregarded the temporal property while most data are closely related time. Moreover, they do not address accuracy degradation due sparsity of time slices. The essential problems exploit and consider slices remain unresolved. In this paper, propose time-aware (tatd), an accurate method tensors. tatd is designed dependency time-varying We new smoothing regularization with Gaussian kernel modeling dependency. improve performance by considering sparsity. design alternating optimization scheme suitable our regularization. Extensive experiments show that provides state-of-the-art decomposing