FacIt: Factorizing Tensors into Interpretable and Scrutinizable Patterns

Abstract

Tensor Factorization has been widely used in many fields to discover latent patterns from multidimensional data. Interpreting or scrutinizing the tensor factorization results are, however, by no means easy. We introduce FacIt, a generic visual analytic system that directly factorizes tensor-formatted data into a visual representation of patterns to facilitate result interpretation, scrutinization, information query, as well as model selection. Our design consists of (i) a suite of model scrutinizing and inspection tools that allows efficient tensor model selection (commonly known as rank selection problem) and (ii) an interactive visualization design that empowers users with both characteristics- and content-driven pattern discovery. We demonstrate the effectiveness of our system through usage scenarios with policy adoption analysis.