Intelligent interfaces for preference-based search

Abstract

Preference-based search, defined as finding the most preferred item in a large collection, is becoming an increasingly important subject in computer science with many applications: multi-attribute product search, constraint-based plan optimization, configuration design, and recommendation systems. Decision theory formalizes what the most preferred item is and how it can be identified. In recent years, decision theory has pointed out discrepancies between the normative models of how people should reason and empirical studies of how they in fact think and decide. However, many search tools are still based on the normative model, thus ignoring some of the fundamental cognitive aspects of human decision making. Consequently these search tools do not find accurate results for users. This tutorial starts by giving an overview of recent literature in decision theory, and explaining the differences between descriptive, and normative approaches. It then describes some of the principles derived from behavior decision theory and how they can be turned into principles for developing intelligent user interfaces to help users to make better choices while searching. It develops in particular the issues of how to model user preferences with a limited interaction effort, how to support tradeoff, and how to implement practical search tools using the principles.