First Principles Selection of Social Media Visualizations

As the importance of social networks skyrockets, so does the need to understand human behavior in these forums. While statistical analyses can help to some extent, many types of patterns are simply not visible by looking at raw numbers. However, many of these patterns do emerge in visualizations, and the innate human ability to see these patterns in graphical depictions represents a valuable opportunity for analysts. Yet the state of social network visualizations is in need of more creative approaches, beyond all-too-common node-link diagrams, which can often result in"hairball"visualizations that do not result in any useful insights. To address this need, we propose to build a fully integrated and deployable version of SocialViz, a first-principles powered visualization system we prototyped in Phase I of this STTR. SocialViz generates a large space of possible visualizations by combining various data properties in a multitude of ways, and then rely on cognitive first principles and human interest models to filter and sort through candidates for visualizations that are predicted to have high utility. SocialViz builds upon cognitive first principles, image analysis techniques, and recommender systems to deliver visualizations that are"interesting", both from a purely cognitive point of view, as well as from the view of that user. In building SocialViz, we will implement a cognitive measures model, a user interest model, and then combine these technologies in a stand-alone system that will be assessed by social network scientists as well as operational-minded subject matter experts. Based on their input, we will iteratively refine our system. Our overarching goal in this proposal is to engage human pattern detection ability to give us insight into social network behavior and to lead us to ask questions that we would have never asked when simply presented with statistics.