[2D6-GS-2] Machine learning: Bayesian estimation

[2D6-GS-2-03] Variational Bayesian Logistic Factorization Machines

〇Yoichi Kitahara¹ (1. Livesense Inc.)

Keywords:collaborative filtering, bayesian estimation

Factorization Machines (FM) are a generalized model of Matrix Factorization (MF) that enables the utilization of side information in collaborative filtering. In cases where the evaluation is binary, logistic regression-type models are commonly used. While FM is a highly expressive model capable of representing many MF-derived models, it tends to overfit the training data and suffers from high computational complexity.In this study, we propose a parameter estimation method for logistic regression-type FM that addresses the issues of overfitting and computational complexity. This method utilizes variational Bayesian inference to suppress overfitting and precomputation to reduce computational quantity.