Author(s)

B Arslan, F Ricci, N Mirzadeh & A Venturini

Abstract

The main objective of a personalized recommender system is to filter and present (recommend) to the user the most appropriate items according to his preferences. In many Case Based Recommendation systems, this goal is achieved by using weighted similarity measures. Thus, weighting the features, i.e. describing the items to be recommended, is a key issue in such systems. In this paper, we propose a dynamic weighting scheme for a Case Based Recommendation System, which is based on statistics of data extracted from past sessios. The applications of these ideas to an interactive, Case-Based travel recommender system, called Dietorecs1, that guides European travelers for their travel decision making processes, are described. 1 Introduction Case-Based Reasoning (CBR) systems, when confronted with a new problem, retrieve a set of similar problems from their case base, which provide a set of viable and potentially useful solutions to the new problem. In case these solutions do not fit the current situation as desired, that is if reuse of these solutions is not suitable, then they can be adapted to the new problem. The final solution, together with its original problem specification, forms a new case to be stored in the case base to \“improve” future performance of the system. CBR is a cognitively appropriate approach for modeling and explaining human problem solving [9], especially in domains where experiences play an important role. Thanks to these advantages, this technique has been gaining popularity in advisory systems [4, 7]. 1This work has been partially funded by CARITRO foundation (under contract \“eCommerce e Tur-ismo”) and by the European Union’s Fifth RTD Framework Programme (under contract DIETORECS IST-2000-29474).

Keywords