Next: Genetic Algorithm Up: The Learning Module Previous: The Learning Module

User Feedback

User feedback has an effect at two levels. One is at the population level, where the fitness of members of the population increases or decreases depending on the feedback. This is required for genetic learning. The other effect is on the profile, which is modified based on the features of the article the feedback is given for. The user can also program the agent by demonstrating examples of documents and providing feedback for those.

As will be clear in the description of the Information Filtering Module, the document representation is generated while scoring the document and is not stored thereafter. Therefore, when feedback for a retrieved document is provided, its representation is no longer available for modifying the profile. This problem is much more acute in programming by demonstration, where feedback is provided for an article for which no representation has yet been created. The solution adopted in this implementation is that the pointers of documents are stored when feedback is provided, but the profiles are not immediately modified. The list of items labeled ``ArtFeedback'' in table is the list of pointers to articles which received feedback in the last user session. Next to the pointers is a positive or negative number indicating the feedback for the article. The next time the Information Filtering module filters articles, it re-creates the document representation for all articles and the profile is then modified.


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