Beagle : automated extraction and interpretation of visualizations from the web

Abstract

In this paper, we present Beagle, an automated data collection system to mine the web for SVG-based visualization images, label them with their corresponding visualization type (i.e., bar, scatter, pie, etc.), and make them available as a queryable data store. The key idea behind Beagle is a new SVG-based classification design to more effectively label visualizations rendered in a browser. Furthermore, Beagle is designed from the ground up to be extendable and modifiable in a straightforward way, to anticipate when new artifacts appear on the web, such as new JavaScript libraries, new visualization types, and better browser support for SVG. We evaluated Beagle's classification techniques on multiple collections of SVG-based visualizations extracted from the web, and found that Beagle provides a significant boost in accuracy compared to existing classification techniques, across a wide variety of visualization types.