The results of the search process was a matrix of raw cocitation data for each pair of URLs in the core set items. The matrix is shown in Table 2. The raw cocitation matrix was entered into the SAS system for the further analyses. It was first converted to a correlation matrix. The purpose of this step was to change raw co-occurrence information to values that can be treated as a form of topical proximity between WWW sites. Sites with high correlations can be seen as similar to one another.

The next step was to process the correlation matrix using the SAS Multidimensional Scaling procedure (MDS). Multidimensional Scaling is a class of techniques for uncovering ``hidden structure'' in databases. The result of applying MDS to a set of proximities results in the output of a spatial representation of the data consisting of a geometric configuration of points, in effect a mapping of the information onto a two or three dimensional plane.

Mon Jul 29 09:00:12 PDT 1996