Decision-theoretic optimization is becoming a popular tool in the user interface community, but creating accurate cost (or utility) functions has become a bottleneck --- in most cases the numerous parameters of these functions are chosen manually, which is a tedious and error-prone process. This paper describes ARNAULD, a general interactive tool for eliciting user preferences concerning concrete outcomes and using this feedback to automatically learn a factored cost function. We empirically evaluate our machine learning algorithm and two automatic query generation approaches and report on an informal user study.

Abstract not found

We propose new methods of preference elicitation for constraint-based optimization problems based on the use of minimax regret. Specifically, we assume a constraintbased optimization problem (e.g., product configuration) in which the objective function (e.g., consumer preferences) are unknown or imprecisely specified. Assuming a graphical utility model, we describe several elicitation strategies that require the user to answer only binary (bound) queries on the utility model parameters. While a theoretically motivated algorithm can provably reduce regret quickly (in terms of number of queries), we demonstrate that, in practice, heuristic strategies perform much better, and are able to find optimal (or near-optimal) configurations with far fewer queries.

The quality of the results produced by personalized e-service applications like product recommenders, buying advisory applications, or product configurators is strongly determined by the accuracy of the system’s estimate of the individual customer’s real needs and preferences. In particular in domains where customers cannot be classified automatically, e.g., based on past buying behavior, these needs have to be interactively elicited by questioning the user. In many existing systems only a “one-style-fits-all ” approach based on static fill-out forms is chosen. However, this does not take the user’s background or capabilities into account, which consequently leads to a poor quality of the acquired user model. In this paper, we show how extensive personalization of the user preference elicitation process itself can significantly improve the accuracy of interactively acquired user models. A comprehensive view on adaptation and personalization opportunities in the elicitation process is developed and corresponding examples for the domain of interactive buying advisory are given. The presented personalization and adaptation techniques are implemented in a domain-independent software framework for building interactive advisory applications. We describe specific architectural requirements for such a system and discuss results from various real-world applications. 1.

Abstract not found

Abstract. While the growing number of learning resources increases the choice for learners, it also makes it more and more difficult to find suitable courses. Thus, improved search capabilities on learning resource repositories are required. We propose an approach for learning resource search based on preference queries. A preference query does not only allow for hard constraints (like ’return lectures about Mathematics’) but also for soft constraints (such as ’I prefer a course on Monday, but Tuesday is also fine’). Such queries always return the set of optimal items with respect to the given preferences. We show how to exploit this technique for the learning domain, and present the Personal Preference Search Service (PPSS) which offers significantly enhanced search capabilities compared to usual search facilities for learning resources. 1

Abstract. We study information elicitation mechanisms in which a principal agent attempts to elicit the private information of other agents using a carefully selected payment scheme based on proper scoring rules. Scoring rules, like many other mechanisms set in a probabilistic environment, assume that all participating agents share some common belief about the underlying probability of events. In real-life situations however, the underlying distributions are not known precisely, and small differences in beliefs of agents about these distributions may alter their behavior under the prescribed mechanism. We propose designing elicitation mechanisms in a manner that will be robust to small changes in belief. We show how to algorithmically design such mechanisms in polynomial time using tools of stochastic programming and convex programming, and discuss implementation issues for multiagent scenarios. 1

Abstract. This paper discusses the design and development of a preference-based search tool (PBST) for tourists, operating on PDA devices. PBSTs are decision support systems that help users in finding the outcomes (e.g., multi-attribute products or services) that best satisfy their needs and preferences. Our tool is specifically aimed at filtering the amount of information about points of interest (POIs) in a geographic area, thus supporting users in the search of the most suitable solution to their needs (e.g., a hotel, a restaurant, a combination of POIs satisfying a set of constraints specified by the user). We focus on the design of an effective interface for the tool, by exploring the combination of dynamic queries to filter POIs on a map with a visualization of the degree of satisfaction of constraints set by the user. We also report the results of a usability test we carried out on the first prototype of the system. 1

Users often have to search for a most preferred item but do not know how to state their preferences in the language allowed by the system. Example-Critiquing has been proposed as a mixed-initiative technique for allowing them to construct their preference model in an effective way. In this technique, users volunteer their preferences as critiques on examples. It is thus important to stimulate their preference expression by the proper choice of examples, called suggestions. We analyze what suggestions should be and derive several new techniques for computing them. We prove their effectiveness using simulations and live user studies.

While trying to satisfy a user’s preferences for a resource, partially stated or completely unknown preferences can be very disrupting. Unfortunately most of the time a user will not specify her preferences perfectly, and her partially stated or unknown preferences will be so numerous that she will not be willing to provide clarifications for all. To complicate things further we may not have perfect information about the resource as well and while we can ask an information source about the resource, the problem of “too many potential questions ” remains if we were to expect getting answers for all the imperfectly specified properties. In cases like these, we need a mechanism for figuring out which questions would yield a bigger improvement in allocating resources to users so that we can ask as few questions as it is possible to answer and get as high of a improvement as possible An example of such a problem is optimization for calculating the best assignment of a set of rooms to a set of sessions. The assignment depends on the various properties of rooms as well as the requirements for each session and the properties of these