User-Guided Interactive Graph Layout

Current graph visualization techniques gather information found in a database or a file and create output on the screen that represents that data. There are different types of layouts from which to choose, such as force-directed layouts, circle-layouts and spring layouts. These layouts are all either random, or based upon the edges in the graphs. They do not lend any weight to what the nodes might represent. By making the graph interactive, thereby allowing the user to manipulate the graph, the graph slowly develops into the organization which the user intends. However, with this method, the user would have to rearrange every node in order to accomplish the intended arrangement. This is inefficient and frustrating to the user. If we incorporate the attributes into the arrangement of the nodes, we should be able to arrive at the correct arrangement much faster than without taking the attributes into account. This may be accomplished by various means, but the two that we considered for this research are simple clustering and constrained clustering. Just performing simple clustering, by using an algorithm such as k -Means, does allow the graph to take attributes into consideration, but does not allow user feedback. By employing an algorithm such as COP-KMeans or PCKMeans, we take both attributes and user feedback and use them to arrange the on-screen graph. We propose that combining a user-friendly interface with some type of constrained clustering will allow the user to arrive at the desired graph in a significantly shorter amount of time.