Using a landscape metaphor to represent a corpus of documents

Abstract

In information retrieval, sets of documents are stored and categorised in order to allow for search and retrieval. The complexity of the basic information is high, with representations involving thousands of dimensions. Traditional interaction techniques therefore hide much of the complexity and structure of the modelled information, and offer access of the information by means of isolated queries and word searches. Bead is a system which takes a complementary approach, as it builds and displays an approximate model of the document corpus in the form of a map or landscape constructed from the patterns of similarity and dissimilarity of the documents making up the corpus. In this paper, emphasis is given to the influences on and principles behind the design of the landscape model and the abandonment of a ‘point cloud’ model used in an earlier version of the system, rather than the more mathematical aspects of model construction.