SetCoLa: High-Level Constraints for Graph Layout

Jane Hoffswell, Alan Borning, and Jeffrey Heer

Paul G. Allen School of Computer Science & Engineering, University of Washington

Summary

    This work proposes a high-level constraint language for graph layout, especially domain-specific graph. In domain-specific graph data, the layout not only considers the topology structure of data, but also employ domain-specific knowledge. Users have the requirements to adjust the graph layout according to nodes properties. It’s constraint. SetCoLa ease users’ burden to set such constraints on graph layout. User can group nodes according to their attributes or local structures. Then different constraints can apply to the set of nodes, like alignment, order, particular layout way. For a set, multiple constraints can be added.

Comments

Strength

    A quite interesting and solid work, which has great application scenarios. Like D3 or Vega, SetCoLa can help users easily create graph layouts. How to create specific or constrained graph layout is pain point in graph visualization. We all encounter this problem during coding. For example, in DBLPVis, we set different length between different categories of nodes. It is not convenient to use D3 to do such configurations. But, we accepted the thinking pattern, without further considering how to solve this problem. Need to reflect everyday life or habit if we can have a method to improve it. As the saying goes, the lazy people make creations because they don’t want to speed much time on their work. 

Weakness

    A further consideration is how to generate the constraints because it’s not easy to explicitly translate your ideal graph layout into clearly mathematical expressions. We have the following points.

    1. Generate constraints from local interactions. Then these constraints can be applied to global. We need to design a method to understand and explicitly list users constraints from their interactions or modifications on graph layout so that users don’t need to generate constraints by themselves especially when they don’t know about programming.

    2. Generate constrains from final layout. Given one constrained layout and an original layout, is it possible to extract the constraints from these two results?

    3. Evaluation of the constrained layout. Such evaluations need much more domain knowledge, which is different from a general graph layout limited to aesthetic metrics.