Learning Subjective Functions with Large Margins

In manyoptimization and decision problems the objective function can be expressed as a linear combinationof competingcriteria, the weights of whichspecify the relative importanceof the criteria for the user. Weconsider the problem of learning such a "subjective" function frompreference judgmentscollected from traces of user interactions. Wepropose a newalgorithm for that task based on the theory of Support Vector Machines. Oneadvantage of the algorithm is that prior knowledgeabout the domaincan easily be included to constrain the solution. Wedemonstrate the algorithm in a route recommendation system that adapts to the driver's route preferences. Wepresent experimental results on real users that showthat the algorithm performswell in practice.

• Type: Conference
• Year: 2000
• Where: ICML
• Authors: Claude-Nicolas Fiechter, Seth Rogers
• Keywords: Algorithm Performswell, ICML 2000, Machine Learning, Route Recommendation, Support Vector Machines