Personalized recommendation based on behavior sequence similarity measures

Abstract

Personalized recommendation is attracting more and more attentions nowadays. There are many kinds of algorithms for making predictions for the target users, and among them Collaborative Filtering (CF) is widely adopted. In some domains, a user's behavior sequences reflect his/her preferences over items so that users who have similar behavior sequences may indicate they have similar preference models. Based on this fact, we discuss how to improve the collaborative filtering algorithm by using user behavior sequence similarity. We proposed a new Behavior Sequence Similarity Measurement (BSSM) approach. Then, different ways to combine BSSM with CF algorithm are presented. Experiments on two real test data sets prove that more precise and stable recommendation performances can be achieved. © Springer International Publishing Switzerland 2013.