LinkScope: Interactive Graph Analysis of Unstructured Text

This paper presents LinkScope, a toolkit for interactive analysis of text using node link graphs, with support for dynamic addition of attributes from tabular data. The interaction technique draws on ideas from 3D modeling, mesh deformation, and static graph drawing to promote discovery of hidden information across a wide variety of graph types and analysis tasks. The key innovation of this work is the application of methods traditionally reserved for automated graph layout and clustering, to produce useful task-specific layout through dynamic interactions. Graph nodes are dynamically repositioned using an interpolated decay function over a single node movement provided by a user. We describe several variants of the interpolation method, including coupling it with a fast local-cut algorithm for cluster selection. Compared to traditional layout mechanisms the technique is particularly useful when meta-data nodes are added to a graph, increasing its connectivity. We show how the techniques can be used interactively to solve text analysis tasks including a case study on a collection of 16K awarded NSF grant proposals with metadata and a corpus of New York Times news articles.