From CS294-10 Visualization Fa08

Jump to: navigation, search



Group Member

Scott Murray

Project Description

Networks, fundamentally, are groups of relationships, whether computer, human, or otherwise. Most network visualizations, however, visually overemphasize the importance of network members while failing to illustrate the nature of the relationships between those members. Simple lines with uniform weight are common, but reveal only that two members are connected. New visualization methods are needed to reveal the qualities of the connections. By visualizing not just connections, but properties of the relationships between members, we could see which relationships are balanced, lopsided, one-way, or reciprocal, and identify similarities and differences in relationships.

A relationship-centric network visualization would de-emphasize a network's members in favor of data on relationships between them. Data sets that indicate connections only could not be used with this new method. Instead, we would need data that describes quantitative values such as:

  • Number of links from one user's blog to another
  • Dollar amounts of contributions made to different organizations
  • Number of emails sent/received between users

Relationships are rarely equal, and a new method could visualize the differences between weak, strong, positive, or negative ones.

Initial Problem Presentation

Problems with Traditional Network Visualizations

  • Placement of nodes is either random or force-directed. Node position on x/y/z planes does not correspond to anything in the initial data set, leaving the connecting lines to carry the burden of information-carrying for the display.
  • Connecting lines are often uniform and, for that matter, only simple lines with no meaningful or defining characteristics. Simple lines are easy to draw, but are two-dimensional, and (without adding properties like color, thickness, opacity, and so on) cannot represent information about the relationship other than whether or not one exists.
  • In areas of dense overlap, connecting lines carry too much visual weight and cannot be followed from end to end. In light areas, each line carries too much weight. Since node placement is random, the angles and length of connecting lines is meaningless and mostly serves to clutter the display.

Possible Solutions

  • Use connecting lines, but add properties such as thickness and color to indicate balance direction.
  • Instead of random or force-directed placement, use spatial relationships to indicate nodes' closeness/distance.
  • Use non-line shapes (perhaps elongated triangles?) to indicate balance direction.

Related Work

  • Adam Perer, Ben Shneiderman. Improving Interactive Exploration of Social Networks. (2006)
  • Adam Perer: Making sense of social networks. Extended Abstracts of ACM conference on Human factors in computing systems. (2006)
  • Adam Perer, Ben Shneiderman. Orderly Analysis of Social Visualizations. (2006)

Useful References

  • Fry, Ben. Visualizing Data

Anticipated Technical Challenges

  • Generating or locating data of appropriate complexity for the project
  • Implementing algorithms to organize nodes and create visuals for relationships between nodes

Implementation Milestones

  • 10/31 — Identify initial data set
  • 11/7 — Complete initial network visualization with nodes and connections
  • 11/21 — Complete refinement of visual elements illustration relationships
  • 12/10 — Complete final paper and poster

Mid-Point Presentation

Final Deliverables

[add comment]
Personal tools