A Summarization-Based Pattern-Aware Matrix Reordering Approach

Matrix-based graph visualization is effective in revealing relationships among entities in graphs. The visibility of structural patterns depends on the ordering of rows/columns in matrices. Most existing approaches mainly settle on an ideal ordering according to quality metrics, which emphasize certain types of patterns but ignore others. This paper proposes a summarization-based pattern-aware reordering approach to highlight multiple patterns simultaneously. First, a pattern-aware graph summarization utilizes the Minimum Description Length (MDL) technique to identify various types of patterns from the input graph. Second, we propose a coarse-to-fine reordering mechanism to generate matrix-based visualizations that maintain the structure of all identified patterns. Experimental results of two comparative studies and a user study on several datasets demonstrate that our approach simultaneously highlights more types of patterns than other approaches and performs well across multiple quality metrics.