Examining collaborative filtering algorithms for clothing recommendation in e-commerce

Abstract

With the boom in online clothing e-commerce, various web portals and mobile applications apply recommendation methods to improve the sales and consumer satisfaction based on the massive historical records in the big data era. This study examined the collaborative filtering algorithms embedded in the typical recommendation methods for online clothing. The test dataset is constructed with a real-world large-scale instance from one of the largest business-to-consumer e-commerce platforms (www.taobao.com) in China. Considering the purchasing times and the inverse user frequency, three similarity measures are developed for the cosine-based similarity algorithm. Various numerical experiments are conducted to analyze the recommendation methods and evaluate their performances by three criteria, namely Precision, Recall and Diversity. Because the test instance is large scale, the consumer–goods co-occurrence matrix is reduced to improve the computational performance considering most similar consumers. Using this real-world instance, the algorithms are investigated under the evaluation criteria. The experimental results reveal that the recommendation based on user frequency similarity is very much suitable for the online clothing recommendation; the co-occurrence matrix reduction method is effective to improve the recommendation performance. To verify the proposed methods, the MovieLens dataset is used for comparison. The results show that the proposed method is suitable for sparse co-occurrence matrices. In addition, limitations, managerial implications and future research directions are also discussed.