Next: Learning from Feedback Up: Filtering Documents Previous: Scoring Documents

Selecting Documents

Each profile searches a part of the database and scores all the available documents. The documents which are finally presented to the user are selected from among the documents which score well with respect to the different profiles. A parameter that can be used while selecting documents, in addition to the document score, is the fitness associated with the profiles. The fitness is a measure of how successful the profile has been in the past in meeting user requirements. There are two variables which can be manipulated while deciding upon the final documents. One is the number of documents and the other is the ordering of the documents. There are a number of approaches to selecting the final documents:

The basic problem is that the ``goodness'' of a document is some function of the fitness of the profile as well as the document score. While there are merits to each of the above approaches, there is no winning approach which would be universally applicable to all users and situations. The right choice could well be a combination of one or more of the approaches and would depend on the user preferences and the kind of documents being filtered.

The reason why a scoring threshold might be inappropriate is that a threshold does not have much meaning in itself. The document scores are good for relative comparisons, but do not have an intrinsic absolute value. Therefore, the recommended approach is a combination of the first two approaches so that both the number of documents and the ordering depend on the fitness.



Next: Learning from Feedback Up: Filtering Documents Previous: Scoring Documents


MIT Media Lab - Autonomous Agents Group - agentmaster@media.mit.edu