Recommendation System With Hierarchical Recurrent Neural Network for Long-Term Time Series

Nowadays, recommendation systems are widely used on various services. This system predicts what item a user will use next using large amounts of stored user history. Recommendation systems are commonly applied in various fields as movies, e-commerce, and social services. However, previous researches on recommendation systems commonly overlooked the importance of item usage sequence and time intervals of the time series data from users. We provide a novel recommendation system that incorporates these temporal properties of the user history. We design a recurrent neural network (RNN) model with a hierarchical structure so that the sequence and time intervals of the user's item usage history can be considered. The model is divided into two layers: a layer for a long time and a layer for a short time. We conduct experiments on real-world data such as Movielens and Steam datasets, which has a long-time range, and show that our new model outperforms the previous widely used recommendation methods, including RNN-based models. We also conduct experiments to find out the influence of the length and time interval of sequences in our model. These experimental results show that both sequence length and time interval are influential, indicating that it is important to consider the temporal properties for long-term sequences.