Applying MapReduce to learning user preferences in near real-time

Abstract

When computer programs participate in conversations, they can learn things about the people they are conversing with. A conversational system that helps a user select a flight may notice that a person prefers a particular seating arrangement or departure airport. In this paper we discuss a system which uses the information state accumulated during a person-machine conversation and a case-based analysis to derive preferences for the person participating in that conversation. We describe the implementation of this system based on a MapReduce framework that allows for near real-time generation of a user's preferences regardless of the total case memory size. We also show some preliminary performance results from scaling tests. © 2013 Springer-Verlag.