FOA Home | UP: Spreading activation search

\RelFbk in AIR

Queries subsequent to the first are performed much differently. After AIR is done retrieving the network of features, the user responds with indicating which features are considered (by that user) relevant to the query and which are not. Using a mouse, the user marks features with the symbols: {\bf ++, +, --} and {\bf -- --}, indicating that the feature was {\tt Very Relevant, Relevant, Irrelevant}, or {\tt Very Irrelevant}, respectively. Not all features need be commented upon.

The system constructs a new query directly from this feedback. First, terms from the previous query are retained. Positively marked features are added to this query, as are the negated versions of features marked negatively. Equal weight is placed on each of these features, except that features marked {Very Relevant} or {\tt Very Irrelevant} are counted double.

From the perspective of retrieval, this RelFbk becomes a form of {\em browsing}: positively marked features are directions which the user wants to pursue, and negatively marked features are directions which should be pruned from the search. From the perspective of learning, thisRelFbk is exactly the training signal AIR needs to modify its representations through learning. This unification of learning (i.e., changing representations) and doing (i.e., browsing) was a central component of AIR's design. It mean that the collection of feedback is not an onerous, additional task for the user, but a natural part of the retrieval process.

Top of Page | UP: Spreading activation search | ,FOA Home

FOA © R. K. Belew - 00-09-21