The Hierarchical Model for News Recommendation

Abstract

Latent factor model (LFM) is a classical model-based collaborativefiltering approach that explains the user-item association by characterizing both items and users on latent factors inferred from rating patterns. Due to high data volume, we consider whether it is reasonable that LFM is a linear model with interaction between users and items. Therefore, we propose a hierarchical model, which groups items or users by item features or user features, and establish a LFM on each class. To avoid overfitting and comparison, we use different kinds of regularization to punish the model due to different meanings of regularization. Then we apply the hierarchical model to news recommendation. We use latent dirichlet allocation (LDA) to analyse news, which gets the topic distribution of news that can be used as news feature. The experimental results show the superiority of the hierarchical model in the news recommendation system with higher accuracy than other algorithms.