From paragraph networks to document networks

We investigate hypertext repositories with automatically generated hyperlinks among paragraphs. The algorithms that generate the links yield results with significant common characteristics despite large differences among the algorithms. Furthermore, different repositories, generated by the same algorithms show significant common characteristics, although generated from the return lists of different search engines. Finally, repositories generated for different domains also have common characteristics. This suggests that these characteristics are pervasive properties of judiciously retrieved document sets. The common characteristics revolve around a tendency for some documents (the kernel) to be destinations of a disproportionately high fraction of the hyperlinks in the repository. An emergent property of such repositories with practical significance is that browsing activities will have a tendency to trap the user within the kernel, perhaps without their realizing it. To enable users to choose to avoid trapping, we propose a ring structure for hypertext repositories which is exposed to the user by annotations to the hyperlinks.