Textual Influence Modeling Through Non-Negative Tensor Decomposition