We define the user profile to be the set of words which tend to partition the user's document accesses into groups. These are words that have occurrence patterns such as that shown in figure 1. Once WordSieve has experienced a reasonably diverse set of document accesses, the user profile can be extracted from the nodes of levels 2 and 3. The user profile consists of the set of words corresponding to nodes in those levels, each word associated with the product of its excitement values in levels 2 and 3. Words with high products tend to have the desired occurrence patterns.
Note that this user profile will be different for every user, because it depends on the user's patterns of document access. These profiles enable the system to learn about the kinds of documents the user accesses as a group, and use them to identify terms which indicate the context in which a user is working. Having identified the context, the system can index a document by that context for suggestion later, and can identify previously indexed documents which may be helpful to the user, aiding personalization of recommendations.