Persistently Effective Query Selection in Preference Elicitation

The selection of queries that will provide maximum information regarding a user´s preferences is a key component of effective preference elicitation. We discuss a technique for selecting a candidate set of comparison queries whose answers will reveal a significant amount of information about the user´s preferences. Computationally expensive utility evaluation of queries can then be confined to this set. Furthermore, this set of queries is chosen so that the response to one query does not resolve any other queries in the set, thus eliminating the need to recompute a new candidate set each time. Experiments run on a case with 30 outcomes show that our chosen queries reveal two to three times as many preferences as random selection, and asking our persistent set of queries reveals 10-12% more preferences than the best n individual queries.