|
|
|
|
|
|
Help | Seminars List | Add Seminar | Edit Seminars | Tips for organisers | RSS | ics Calendar | Search | Send comments about this website to seminar-master@cecs.anu.edu.au
Contact: Michelle.Moravec@anu.edu.au CS PHD MONITORING
Value of Information Heuristics for Bayesian Preference ElicitationMr Shengbo Guo (School of Computer Science, CECS)DATE: 2009-09-24 TIME: 10:45:00 - 11:15:00 LOCATION: RSISE Seminar Room, ground floor, building 115, cnr. North and Daley Roads, ANU ABSTRACT: The aim of preference elicitation (PE) is to efficiently elicit a user's preferences in order to maximize her utility for a given task. There are three key aspects of PE: (1) modeling the user's latent utility function, (2) formulating a utility elicitation strategy, and (3) making a recommendation under uncertainty that minimizes the user's expected utility loss. In this work, we model the belief in the user's latent utility function using a normal distribution. Considering the system's belief over the user's utility function as prior, we use expectation propagation (EP) and assumed density filtering to update the utility beliefs based on the user's response to preference queries. This probabilistic framework naturally facilitates the use of value of information as an efficient elicitation strategy for query selection. Experimental results on a simulation data set demonstrate that our approach consistently outperforms other elicitation strategies.
|