A Personalization Technique using Multiple Decision Trees in Recommender System

D. Lee and S.J. Hyun (Korea)


Data mining, Decision Tree, Ecommerce, Personalization, Recommender system


In the e-commerce and online services, personalization techniques have been employed in a wide variety to provide a higher level of customization. Most of recommendation systems are based on the tendency of the group with similar attributes. However, users very often show some unique characteristics that are different than the general tendency. According to the business policy, it may be necessary to take into account some particular attributes of individuals. In this paper we propose an attribute-based recommendation algorithm to reflect both the general tendency and the individual preferences in a way modeling multiple decision trees for both the group and individuals. The algorithm uses the weight factors which represent the unique preferences of the active users on some particular attributes. A weight factor is derived from entropy over some particular attributes of an active user. It is used to weigh the amount of reflection of group tendency and individual preference to the recommendation algorithm. We show the modeling of multiple decision trees and data structures, design of the proposed algorithm, and performance analysis of an implemented recommender system. With the proposed personalization technique e-commerce and online services would achieve an additional level of personalization.

Important Links:

Go Back