Association rules and collaborative filtering on sparse data of a leading online retailer
Yongzhong Wu, Mianmian Huang, Yuxin Lu · 2017
Personalized recommender systems are important for online shopping retailers to recommend items to potential customers. However, data sparsity is a key problem leading to poor recommendations. In this paper, we established two recommender models, i.e., the one based on association rules and the one based on collaborative filtering (CF), and tested them on a large set of sparse data obtained from a Chinese leading online shopping retailer. For the first model, only a limited number of reliable association rules were obtained due to data sparsity. For the second model, although collaborative filtering did not work well on the entire dataset, it performed significantly better when limiting the data to those associated with popular items. As retailers' majority of revenues coming from popular items, restricting the dataset to those associated with popular items can improve the effectiveness and maintain the usefulness at the same time.