PivotGraph

A way of visualizing multivariate data that reduces the data into two dimensions and shows the relationships between them

by Kent Ratliff and Emily Sarich

Overview

The PivotGraph is a visualization technique developed by IBM developers in 2006. It takes data, usually representing some sort of connectivity between groups, and aggregates it to show trends and relationships between the groups. This aggregration reduces the number of variables to look at, giving the user a better understanding of general trends.

How it works

The PivotGraph starts with a selection of dimensions in a set of multivariate, connected data. It then uses a rollup algorithm to aggregate the data into discrete categories, as shown below:

Once the data has been aggregated, it assigns each dimension a row or column and places a vertex in the correct row/column for each attribute.

It then draws in the connections between each vertex, with the width of the edge representing the number or strength of the connection between the vertices.

This implementation of the PivotGraph also colors the vertices based on the number of incoming connections versus the number of outgoing connections, and the edges based on the number of connections. It also allows users to select which dimensions to roll up by, and filter down the data, giving a great deal of interactivity that makes it even easier to explore the data:

Evaluation

Advantages:

• The PivotGraph is good for seeing trends in multivariate data.
• It displays clearly the relative strength or size of connections between vertices.
• Because it aggregates data, it's very scalable and good for large data sets.
• It can easily be made to be an interactive visualization, where users can choose which dimensions to display

Disadvantages:

• Though the aggregation and reduction of dimensions is a helpful tool, it does result in some loss of information, as only a limited number of dimensions can be displayed at a time.
• In all current implementations, PivotGraphs rely on discrete-dimensional data, so we can't visualize any continuous datasets.
• By nature of the type of visualization, PivotGraph can only be used for data with some sort of connectivity (email correspondance, social networks, etc.). Therefore, it is limited in its uses.