Personalized recommendation based on weighted sequence similarity

Abstract

The sequential pattern mining-based recommendation has recently become a popular research topic in the field of recommender system. However, this kind of methods usually relies on frequency counts of sequences, which makes low-frequency sequences contribute little for the final recommend results. To solve this problem, in this paper, we propose a weighted sequence similarity-based method, called Personalized Recommendation based on Sequence Similarity (PRSS), for personalized recommendation. First, item-sequence weight model is introduced, which can reflect different importance of different items to different sequences. Then, target users’ sequence is compared with historical sequences using similarity function. Finally, the maximal common subsequence is proposed to rank candidate sequences and make recommendation. Experimental results show that PRSS generates more accurate recommendation for the target users.