Accelerating Sparse Tensor Decomposition Using Adaptive Linearized Representation