Organix: Creating Organic Objects from Document Feature Vectors

2.0 Related Work

A wide variety of document collection visualisations have been developed. Bead (Chalmers & Chiston, 1992) uses physically based modelling techniques to produce document clusters. This approach can be computationally complex, an alternative and more efficient algorithm has been developed (Chalmers, 1996). WEBSOM as described by Lagus et al. (1996) uses a self organising map (SOM) algorithm to produce a map of documents with similar documents located in closely related regions of the map. Themescapes (Wise et al., 1995) visualises the thematic content of a document collection as a 3D landscape, stronger themes are give a higher elevation. A network is used by Singhal and Salton (1995), Salton (1995), the resulting structure of the network and the number of incident lines (or degree) at a particular node can give insights into the core documents or paragraphs within a particular article. The research and approaches used for text visualisation are extensive. Card et al. (1999, p409-461) contains a selection of papers discussing 1D, 2D and 3D text visualisation. A comprehensive review of document visualisation has been written by Morse (1998).

A novel system is described by Roher et al. (1998). This approach generates a document feature vector, maps the weights for each feature to distances along each axis and the eight bisecting quadrants, place spheres at the end points and finally produce a 3D amorphous shape. This allows up to 14 dimensions of the document to be viewed as a single shape. Documents can then be compared, with similar documents having similar shapes. It was this idea that inspired the current work.