AIC Seminar Series

Preference-based search is the problem of finding the most preferred item among a large collection. The most popular methods for preference-based search are based on question answering, and only a minority of people manages to find their most preferred item. In this talk we present new mixed-initiative tools using example-critiquing and suggestions that, according to user studies, dramatically increase decision accuracy.

We also consider the problem of preference-based search for configurable products, where online electronic catalogs can be modeled by Constraint Satisfaction Problems. We show how the generation of the displayed examples can be formulated as a single optimization problem and solved sufficiently fast for practical applications.

Finally we discuss a strategy for producing suggestions that exploit prior knowledge of preference distributions and can adapt the displayed examples to users’ reactions.

Paolo Viappiani is an assistant and PhD candidate at the Artificial Intelligence Laboratory of the Ecole Polytechnique Fédérale de Lausanne (EPFL). His thesis is advised by Prof. Faltings and focus on personalized search tools. His research is conducted in collaboration with the Human Computer Interaction group directed by Prof. Pu, in the same institute. He holds a Master degree (Laurea) in Computer Engineering from the "Politecnico di Milano", Italy and is a member of AAAI and AI*IA (Italian association for Artificial Intelligence). His interests include reasoning about uncertainty, constraints and preferences. More details can be found at http://people.epfl.ch/paolo.viappiani