Improving Topic Model Visualization via Multi-Dimensional Scaling and Cliques

We propose representing high-dimensional data in 2-dimensions using cliques mapped onto several planes. Currently, Multidimensional Scaling (MDS) projects every point onto an R space medium. However, this may not produce the most ideal result as some relations between points may exhibit higher stress than others. We propose utilizing cliques to extract a complete subset of points into separate facets in order to convey the most accurate distance representation as possible, therefore achieveing low stress in each instance.