Visualizing content based relations in texts

Abstract

Our goal is to efficiently visualize a medium sized hypertext database containing 500-20000 articles. The visualization technique we propose is an information landscape. Basically, the information landscape maps texts into a 2D plane so that related texts are placed next to each other. The hypertexts' location is calculated according to their content and not according to their links. Combining already published algorithms, the clustering works very well. An important issue, however, is a well-designed user interface (UI). The article presents two 2D interfaces and an improved 3D version. The paper covers all aspects from preprocessing and clustering to the final UI and its functionality.