Soojung Lee, Journal of Internet Computing and Services, Vol. 26, No. 3, pp. 67-74, Jun. 2025
10.7472/jksii.2025.26.3.67, Full Text:  HTML
Keywords: Recommender System, Collaborative Filtering, Similarity Measure, data sparsity

Abstract

As the amount of information available on the web increases rapidly every day, the introduction of information filtering techniques for user search efficiency is essential. Collaborative filtering is one of the most useful methods for recommending products to e-commerce users. The most important component of this method is to find similarities between users using the user-item rating matrix and recommend products preferred by similar neighboring users. However, one of the key problems is that the system performance deteriorates due to the difficulty in measuring similarity when the user rating data is sparse. In this study, we propose a base similarity measure for sparse data environments such as Jaccard coefficient. The proposed method introduces a semantic approach to rated items to overcome the limitations of Jaccard coefficient by basing it on the number of common rated genres rather than the number of common rated items of two users. The performance evaluation was performed using two types of public datasets, and the results show that the proposed method outperformed the prediction performance and coverage performance of the Jaccard coefficient by up to 7.4% and 35.9%, respectively, and showed excellent prediction performance especially in sparse data environments. In addition, despite not utilizing user evaluations, the proposed method generally outperformed the performance of existing traditional similarity measures.

Statistics
Cite this article