Towards Unambiguous Edge Bundling: Investigating Confluent Drawings for Network Visualization

Benjamin Bach, Nathalie Henry Riche, Christophe Hurter

IEEE Transactions on Visualization and Computer Graphics

|November 23, 2016

Summary

This study explores a new way to visualize networks called Confluent Drawings (CD). Traditional edge bundling can cause confusion because it groups edges based on their positions, not their connections. CD instead bundles edges based on shared sources or targets, which better reflects network structure. The researchers created an algorithm to generate CDs from any network and tested it in a user study. They found that non-experts could understand CDs more accurately than other bundling methods. This suggests that CDs may be a useful tool for clearer network visualization.

Area of Science:

Network visualization techniques in data science
Graph theory applications in computer science

Background:

Network diagrams often suffer from visual clutter due to overlapping lines. Prior research has shown that bundling edges can reduce this clutter. However, existing methods may introduce ambiguity in connectivity. Confluent Drawings (CD) propose a different approach. Unlike traditional bundling, CD groups edges by shared sources or targets. This aligns bundling with network structure rather than spatial proximity. No prior work has tested CD on arbitrary networks. The lack of practical tools limits real-world use. This gap motivated the development of a new algorithm. The study aims to bridge theory and application in edge bundling.

Purpose Of The Study:

The goal is to evaluate CD as a visualization method for networks. The study introduces a new algorithm for constructing CDs from any network. It also compares CD with other bundling techniques. The motivation stems from the need for clearer edge representation. Current methods may mislead users about connections. The authors aim to test if CDs improve connectivity perception. They seek to validate CD's potential through user studies. The focus is on usability for non-expert audiences.

Main Methods:

The researchers developed an algorithm for generating CDs from arbitrary networks. They implemented a layout method within a sandbox environment. This allows interactive exploration of network structures. The study includes a comparison of CD with other bundling techniques. Artifacts and patterns in CDs were analyzed systematically. A user study was conducted to assess readability. Participants included individuals without visualization expertise. The study tested comprehension of connectivity in small networks.

Main Results:

The algorithm successfully generated CDs for arbitrary networks. Interactive tools enabled exploration of network structures. CDs showed fewer perceptual errors than traditional bundling. Users identified correct connections more reliably with CDs. The user study found that non-experts could interpret CDs effectively. Artifacts in CDs were distinct from those in other methods. Power graphs and metro-style bundling had higher error rates. CD's structure-based bundling improved connectivity perception.

Conclusions:

The authors propose that CDs offer a clearer representation of network connectivity. Their algorithm enables practical use of CDs for arbitrary networks. The user study supports the claim that CDs reduce perceptual errors. The sandbox environment facilitates interactive exploration. CDs outperformed other bundling techniques in user tests. The findings suggest that CDs may be more intuitive for non-experts. The study does not claim CDs are universally superior. It suggests that CDs could be a valuable addition to visualization tools.

Keywords:

network visualization techniques edge bundling methods graph drawing algorithms interactive network exploration

Frequently Asked Questions

Confluent Drawing bundles edges by shared sources or targets, while traditional methods use spatial proximity. This reduces perceptual errors in connectivity.

The sandbox environment allows interactive exploration of network structures using the CD algorithm. It supports real-time adjustments and visualization.

Planarity restrictions limit CD application. The study's algorithm works on arbitrary networks, making CDs applicable beyond planar graphs.

The study found that non-experts could interpret CDs effectively. Participants had fewer errors in connectivity perception compared to other bundling methods.

CDs were compared to power graphs, metro-style, and common bundling. Artifacts and user performance were analyzed to assess clarity and accuracy.

The authors propose that CDs may improve network visualization for non-experts. They suggest CDs could be a valuable addition to existing bundling techniques.