This survey focuses on tensor factorization methods based on stochastic models. They are natural extensions of the conventional methods to Bayesian statistics and are easier to introduce prior knowledge. Some of their applications are also given.