On quantitative reasoning about qualitative preferences

Carmel Domshlak
Cornell University

During the last decade, the Internet has made accessible to the masses large databases containing information about products, services, and all other possible objects of interest. In this talk we overview a novel methodology for elicitation and reasoning (under certainty) about user preferences, based on a "user-friendly" obtaining of a user value function for the purpose of sorting a (structured) database or identifying the most desirable item(s) in it. The core of this methodology is a preference elicitation technique based on TCP-nets and pair-wise comparisons between some items in the database. In large, the methodology is based on a hybrid qualitative/quantitative framework, where the user incrementally provides us with some qualitative statements about her preferences, while the system is working with a value function automatically generated from the given qualitative information.