Factorisation machines considering the latent characteristics behind target data

A factorisation machine (FM) is often used as a regression-type prediction model that considers the interactions between features. In recent years, many researchers have focused on the differences between interactions from the perspective of prediction accuracy. In some cases, it can be assumed that there are several groups with different characteristics in terms of the relationships between input and output data for a target population. Therefore, there is a possibility that the effects of interactions will differ between input data groups, which are used to construct latent clusters. However, there has been little research on FMs considering the latent characteristics behind target data. Therefore, this paper proposes a novel type of FM that considers latent data characteristics. The proposed model can express the differences between the interaction effects of latent groups and it is expected to improve the prediction accuracy of FMs for complex problems. Based on a demonstration experiment using the MovieLens dataset, the effectiveness of our proposed model is verified.