[T-VCG 2022] Multi-Level Area Balancing of Clustered Graphs

[uhhyunjoo]

• 다양한 필드의 complex relationships 를 잘 이해하기 위해 clusted graphs 를 laying out 하기 위한 multi-level area balancing techgnique 을 제안한다.
• Clustered graphs 는 주로 attribute-based grouping information 을 이용하여 relationships 를 모델링하는 데 쓰인다.
• 여기서는 여러 수준의 세부 정보에서 Voronoites tessellations 을 사용하여 input screen space을 계층적으로 분할할 것을 제안한다.

[uhhyunjoo]

graph layout 에 대해 space partioning 을 하는 것

• decomposed into several levels in top-down fashion

• 각 레벨에 대한 area balancing 이 고려된다. information complexity or density of the level underneath

• four-level area balancing approach to allocate appropriate space for each vertext within a cluster

• this design decision has been made based on the topological properties of networks

1. Category-level : cluster properties
2. Component-level : connected-component properties
3. Topology-level : the abstract form of sub-networks
4. Detail-level : the detailed sub-networks
   -> space partitioninig

• in practice, each level is computed by a force-based layout followed by a schematization approach for simplifying the sahpes of the contours to accomplish detail-level vertext area balancing

[uhhyunjoo]

graph skeleton g_s \in {G_C, G_M, G_T, G_D} for each of the for levels

• the layout of each graph skeleton is used to guide the positioning of its belonging vertices to their expected position, in order to retain a balanced distribution.

• category-level

• component-level : we drag components sharing some vertexts close to each other and align cells containing subgrtphs

[uhhyunjoo]

• we report the running time
• Table 2 summarizes the properties of our datasets.
• • number of vertexts, edges, clusters, and graph densities
• subscrit D : after vertext duplication
• used same graph density function

1. Measuring space coverage and time complexity

• fragmented empty space is not fully utilized...
• fully usese the screen as preferred
• we inroduce two coverage measures
  • M_N : coefficient of variation of distances of the vertexts to their k nearest neighbors -> to examine if each vertext has equal distances to its neighbors
  • M_V : corresponds to the number of pixels of its corresponding Voronoi cell.
• both measures show the area assigned to each vertext is more balanced in our approach.
• this tendency increases as the data size increasese.

2. Interview with experts in biology

• six domain experts, who are experienced with manualy creating pathway diagrams, and discussed our selected aesthetic criteria and the quality of the results with them.

(1) : explaining how to read visualization (datasets and color coding)