User-level View

The high-level user view I will be using is adapted from the INQUERY perspective [51]. A user has some Information Need. The user begins a session with an IR system by formulating a query , and submitting it to the IR system. The system returns a set of documents . If satisfies the Information Need, then the session is over. Otherwise, the user reformulates to and the cycle continues.

An important factor affecting how the user will interact with the IR system is the type of Information Need the user has. Typically, there are three types of Information Needs that an IR system can handle:

 

Pinpoint

A user needs an answer to a question; typically, there is one ``right'' answer which may be in one or more documents. Returning any document containing the answer satisfies the Information Need.

Exhaustive

A user needs all relevant information regarding the query. Typically, the user's Information Need is met once a certain number of relevant documents are returned.

Future

A user needs all relevant information now and in the future regarding the query. Typically, the Information Need is satisfied when the system returns an optimized query to the user, as opposed to a set of documents. The methods that satisfy this type of Need are often called ``Document Routing'' or ``Routing'' in the literature.

This paper will be concerned with Relevance Feedback as it applies to the Exhaustive types of Information Need. Relevance Feedback is typically not very helpful with Pinpoint Needs; by definition, the Need is satisfied once any relevant document is found, which is also a necessary precondition for any RF technique. The techniques used for Exhaustive and Future tasks are often similar; however, the additional detail necessary to fully explore Future Needs is beyond the scope of this paper.