Visualizing Similarity with Graph Visualizations

Like just about everything else in this book, there’s certainly more than one way to visualize the similarity between items. The approach introduced in this section is to use graph-like structures, where a link between documents encodes a measure of the similarity between them. This situation presents an excellent opportunity to introduce more visualizations from Protovis, an HTML5-based visualization toolkit produced by the Stanford Visualization Group. Protovis is specifically designed with the interests of data scientists in mind, offers a familiar declarative syntax, and achieves a nice middle ground between high-level and low-level interfaces. A minimal (uninteresting) adaptation to Example 7-7 is all that’s needed to emit a collection of nodes and edges that can be used to produce visualizations similar to those in the Protovis examples gallery. A nested loop can compute the similarity between the working sample of Google+ data from this chapter, and linkages between items may be determined based upon a simple statistical thresholding criterion. The details associated with munging the data and tweaking the Protovis example templates won’t be presented here, but the code is available for download online.