7.3.1 Sources of Information

In order to build user profiles, data about the user must be collected. There are a number of methods available for doing this, including the following.

• Data-mining user's directories

  A prototype Java tool has been developed with which the user can generate a 'first cut' of a user profile. The Java tool is pointed at a user directory (typically the PC's 'My Documents' folder). The tool then uses ProSum to extract keywords from all document files, aggregates the keywords across all these document files, and gives a first approximation of the user's profile. Further facilities could be added, such as finding all Word documents authored by the user (from the Word document properties), and giving greater weighting to keywords found in these documents. However, such techniques rely on the appropriate metadata being correct; this is often not the case.

• Data-mining Web-server log files

  This method has an issue surrounding data validity as the user may access Web pages via a proxy server. A user may look at a Web site's pages with some frequency but those pages may be cached in either the local client cache or a network proxy server. Therefore the total hits by that user would not be accurately recorded on the site. Another major problem here is that since each Web server keeps its own server log, and users move...