Loglinear and multidimensional scaling models of digital library navigation

Computer transaction log analysis comprises the modeling of patterns of machine–user transactions. Due to the web-like structure of the Internet, Interactions can be represented in a three-dimensional, origin–destination–time flow/transaction matrix. Loglinear models are used to model these transactions in terms of their time, origin and destination components and to detect the signature and stability of patterns of navigation through an Internet-based digital library. Through the use of metric multidimensional scaling, transaction patterns are then visualized and interpreted. The methods are applied to a 12-month time series of transaction logs taken from the Alexandria Digital Library, a web-based digital library of maps and other cartographic materials. Among other things, results show that transaction patterns do not change significantly despite changes in the user interface and that user training has a significant impact on people’s navigation of the library. # 2002 Published by Elsevier Science Ltd.