Case-based User Profiling in a Personal Travel Assistant

Abstract

[Introduction] In this paper we present an architecture for a Personal Travel Assistant (PTA). We focus on the ability of this PTA to elaborate a user’s travel requirements and evaluate offers. These decisions are based on a User Profile that has the interesting characteristic that it adopts a lazy policy on learning from a user’s past behavior. Information on user behaviour is stored in a case base in the form of a Case Retrieval Net (Lenz, Auriol & Manago, 1998). A Retrieval Net can capture context-sensitive similarity information which is important in this domain. This lazy approach also has the advantages that it is incremental, extendible and allows flexible reuse of the information. In contrast, eager approaches must make assumptions about how the learned knowledge is to be used. This user profiling system is being implemented as part of the Personal Travel Assistance scenario proposed by FIPA (Foundation for Intelligent Physical Agents).