DeepDrawing: A Deep Learning Approach to Graph Drawing

论文传送门
视频
项目主页

作者

香港科技大学

• Yong Wang
• Zhihua Jin （浙江大学）
• Qianwen Wang
• Huamin Qu

微软亚洲研究院

• Weiwei Cui

IBM T.J. 沃森研究中心

• Tengfei Ma

摘要

节点链接图被广泛应用于网络探索。但是，当使用图形绘制技术来可视化网络时，用户通常需要通过比较相应的绘制结果来迭代地调整不同的算法特定参数，以实现所需的视觉效果。这种反复试验的过程通常很繁琐且耗时，特别是对于非专业用户而言。受深度学习技术强大的数据建模和预测功能的启发，我们探索了将深度学习技术应用于图绘制的可能性。具体来说，我们提出使用基于图LSTM的方法将网络结构直接映射到图绘制。给定一组布局示例作为训练数据集，我们对提出的基于图LSTM的模型进行训练以捕获其布局特征。然后，将训练后的模型用于新的网络以类似的样式生成绘制的图。我们以定性和定量两种方式用两种特殊的布局（即网格布局和星形布局）和两种常规类型的布局（即ForceAtlas2和PivotMDS）来评估了该方法。结果证明了我们方法的有效性。我们还对具有20到50个节点的小图进行了时间成本评估。我们进一步报告了所学到的教训，并讨论局限性和未来的工作。

Motivation

Graph drawing has been extensively studied to facilitate the exploration, analysis and presentation of networks!

However, users often need to find a desirable graph layout through trial-and-error:

• Tune different algorithm-specific parameters
• Compare different drawing results.

It is time-consuming and not user-friendly, especially for non-expert users!

Graph Structures -> Graph Drawing Model using Deep Learning

Graph Structures

Graph Drawing Model using Deep Learning

Graph Drawing

Challenges

• Model Architecture
• Loss Function Design
• Training Datasets

DeepDrawing - Model Architecture

Architecture Details:

• BFS-ordering of graph nodes
• Fake edges and real edges
• Bi-directional

DeepDrawing - Loss Function Design

Procrustes Statistic

Evaluations - Quantitative Evaluation

• Procrustes Statistic-based similarity
• Running Speed
• Training Convergence Comparison

Limitations

• Lack Interpretability
• Our current evaluations mainly focus on small graphs with 20 to 50 nodes.
• The performance of DeepDrawing has a dependence on the input node ordering and the structure similarity with the training graphs.