Abstract
We present an approach to the design of personalized recommender systems that integrates content-based methods, collaborative filtering techniques and case-based reasoning while adopting a user-centered perspective. These techniques are employed to support information search and choice processes. In this framework, we developed and tested a system prototype (NutKing) that helps the user to construct a travel plan by recommending attractive travel products or by proposing complete itineraries. In the information search phase, the system aids the user in specifying a successful query that winnows out unwanted products in electronic catalogues and reduces the information overload. This is accomplished through two kinds of query rewriting operators (relaxation and tightening) in a mixed initiative approach. In the choice phase, the search results are sorted according to a case-base similarity metric, which takes into account the similarity between the users’ travel preferences. The aim of this adaptive sorting is to highlight products that are potentially interesting, because they are similar to those selected by other users in an analogous context. The prototype has been empirically evaluated in a pilot study. The results of the pilot evaluation are discussed, with special reference to aspects concerning the usersystem interaction aspects.