Document Type


Subject Area(s)

Computer Science and Engineering


An approach to browsing large chemical reaction databases is presented. The method that is described builds on earlier work in which unsupervised hierarchical classification was used to extract generalizations of reaction classes from reaction databases for use in reaction knowledge bases. The method described in this paper involves classification based on both semantic and topological features. It supports the creation of deep hierarchies in which succeeding levels represent increasing degrees of abstraction. The creation of a hierarchy allows the user to quickly locate interesting items or classes of items by performing a tree traversal as opposed to sequentially scanning a hit list. In addition, the depth of the resulting hierarchy is determined interactively by the user.