Mining Sequential Patterns of Historical Purchases for E-commerce Recommendation

Abstract

In E-commerce Recommendation system, accuracy will be improved if more complex sequential patterns of user purchase behavior are learned and included in its user-item matrix input, to make it more informative before collaborative filtering. Existing recommendation systems that attempt to use mining and some sequences are those referred to as LiuRec09, ChoiRec12, SuChenRec15, and HPCRec18. These systems use mining based techniques of clustering, frequent pattern mining with similarity measures of purchases and clicks to predict the probability of purchases by users as their ratings before running collaborative filtering algorithm. HPCRec18 improved on the user-item matrix both quantitatively (finding values where there were 0 ratings) and qualitatively (finding specific interest values where there were 1 ratings). None of these algorithms explored enriching the user-item matrix with sequential pattern of customer clicks and purchases to capture better customer behavior. This paper proposes an algorithm called HSPRec (Historical Sequential Pattern Recommendation System), which mines frequent sequential click and purchase patterns for enriching the (i) user-item matrix quantitatively, and (ii) qualitatively. Then, finally, the improved matrix is used for collaborative filtering for better recommendations. Experimental results with mean absolute error, precision and recall show that the proposed sequential pattern mining based recommendation system, HSPRec provides more accurate recommendations than the tested existing systems.

This research was supported by the Natural Science and Engineering Research Council (NSERC) of Canada under an Operating grant (OGP-0194134) and a University of Windsor grant.