FCM-Induced Switching Fuzzy Factorization Machine for Collaborative Filtering
Rikuto Daido, Katsuhiro Honda, Seiki Ubukata, Akira Notsu · 2023
Factorization machine (FM) is a powerful tool for collaborative filtering (CF), which constructs a prediction model for personalized recommendation with lower computational costs. This paper considers extension of FM for handling the mixture of different preference tendencies under the switching data analysis concept. Fuzzy c-means (FCM) clustering has been extended to switching data analysis by modifying the FCM clustering criterion with least square-type measures of various data analysis methods. The proposed method tries to construct multiple FM prediction models for fairly representing intrinsic dependencies among users' preferences and items, where fuzzy membership degrees of each user to cluster-wise models are estimated by utilizing FM prediction error measures not only for FM parameter estimation but also for cluster estimation. The characteristic features of the proposed prediction model are demonstrated through numerical experiments using real-world MovieLens benchmark datasets such that it has advantages for revealing multiple FM prediction tendencies having better prediction ability than the conventional single FM model.