Non-Redundant Sequential Association Rule Mining and Application in Recommender Systems

Many modern recommender systems are not suitable for recommending infrequently purchased products such as cars due to lack of user rating data to infrequently purchased products. A big challenge for recommending infrequently purchased products is the lack of data about users' interests. Web log data is an important data resource to derive useful information about users' navigation patterns which in turn can help find users' information needs. In this paper, a novel method Closed Sequence-Sequence Generator Mining (CSGM) is proposed to generate closed sequences and sequence generators for non-redundant sequential rule mining. By applying the proposed method on web logs, we can extract sequential associations among products which reflect users' preference on products. We have conducted experiments on recommending cars based on users' interests generated by utilizing the sequential rules extracted using our method. Our experiments show that by using those rules we can find users' interests more accurately and thus improve the quality of car recommendation compared to the standard matching-based car search. Moreover, by only using the non-redundant rules, the same or even better recommendations can be generated than using the whole set of rules.