Factorizing Markov Models for Categorical Time Series Prediction

During the last decade, recommender systems became a popular class of models for many commercial websites. One of the best state-of-the-art methods for recommender systems are Matrix and Tensor Factorization models. Besides, Markov Chain models are common for representing sequential data problems (e.g. categorical time series data). The item recommendation problem of recommender systems in fact is a categorical time series problem where each user represents an individual categorical time series. In this paper we combine factorization models with Markov Chain models. To increase efficiency of parameter estimation we introduce our generalized Factorized Markov Chain model.