Artificial Intelligence Lab PhD Seminar

2011 Jul 29 at 11:30

DC 2306C (AI Lab)

Preference Elicitation for Plane Travel

Gregory Hines, PhD candidate, David R. Cheriton School of Comp. Sci., Univ. Waterloo

Preference elicitation allows us to understand peoples' utility values for different outcomes. We are specifically interested in outcomes involving some degree of risk or uncertainty. In this talk, I will consider a case study of helping people to plan plane trips. Flying today involves a high risk of delays; in 2008, more than 1 in 5 domestic flights arrived late in the USA. Delays can often cause missed connections and missed connections can, in turn, greatly increase the duration of a trip. With most airlines using a ``hubs and spokes'' model for moving passengers, direct flights are becoming rarer and the need to make connecting flights the norm. Passengers will have different risk tolerances regarding the probability of a missed connection. Highly risk-averse people will want to have more time between connections. While on average this approach may increase the duration of a trip, it will decrease the probability of an exceptionally long trip due to missed connections. A more risk-tolerant person will be willing to book closer connections. This will decrease the expected duration of the trip but increases the probability of missed connections. Our goal is to design a system which can understand a user's risk preference and help them decide on the optimal trip. In this talk I will discuss issues related to this system which are addressed by my thesis. Specifically, I will talk about how to overcome psychological effects which may prevent people from rational reasoning. Since this system would be used by many different people, I will also discuss how to use results from previous users to improve the performance for future users. Finally, since there are other factors such as cost in choosing a trip, I will discuss how to effectively elicit multiattribute preferences. This will be a high level talk.