Adaptive Online Rough C-Means Clustering and Its Application to Collaborative Filtering
Seiki Ubukata, Tomohiro Kawakami, Katsuhiro Honda · 2022 IEEE Symposium Series on Computational Intelligence (SSCI) · 2022
Content recommendation systems found in e-commerce sites and video streaming services are realized by collaborative filtering (CF). In clustering-based CF, clusters composed of users with similar preference pattern are extracted and items with high preference degree in the cluster to which user belongs are recommended. Rough C-means (RCM) is a clustering method that incorporates a perspective of rough set theory into hard C-means (k-means) to deal with the uncertainty inherent in data. Furthermore, online RCM (ORCM) and its application to CF (ORCM-CF) have been proposed for large-scale datasets. ORCM introduces online learning to RCM and loads an object one by one and gradually updates the parameters. In RCM and ORCM, although it is necessary to set the number of clusters in advance, it is difficult to set the appropriate number of clusters. In this study, we propose adaptive ORCM (AORCM), which aims to automatically generate the appropriate number of clusters by adaptively adding and merging clusters during online learning. Furthermore, we propose AORCM-CF as its application to CF. We verified the recommendation performance of the proposed AORCM-CF through the numerical experiments using real-world datasets.