How CF-NADE Revolutionizes Collaborative Filtering with Neural Autoregression

Recently, Hulu's team had a paper titled "A Neural Autoregressive Approach to Collaborative Filtering" accepted as an oral presentation at ICML 2016. The work applies deep learning to the core problem of collaborative filtering in recommender systems, achieving state‑of‑the‑art results on public datasets such as MovieLens and Netflix.

ICML (International Conference on Machine Learning) is one of the two top conferences in machine learning, alongside NeurIPS. This year ICML received 1,327 submissions, accepted 322 papers (24.3% acceptance rate), and will be held on June 19 in New York.

The first author, Dr. Zheng Yin, joined Hulu's recommendation team in 2015, focusing on recommendation algorithms, computer vision, and deep learning. He has published in TPAMI, IJCV, CVPR and serves as a reviewer for major conferences and journals.

Abstract

This paper proposes CF-NADE, a neural autoregressive architecture for collaborative filtering (CF) tasks, which is inspired by the Restricted Boltzmann Machine (RBM) based CF model and the Neural Autoregressive Distribution Estimator (NADE). We first describe the basic CF-NADE model for CF tasks. Then we propose to improve the model by utilizing rating‑invariant information and sharing parameters between different ratings. A factored version of CF‑NADE is also proposed for better scalability. Furthermore, we take the ordinal nature of the preferences into consideration and propose an ordinal cost to optimize CF‑NADE, which shows superior performance. Finally, CF‑NADE can be extended to a deep model, with computational complexity increased moderately. Experimental results show that CF‑NADE with a single hidden layer outperforms the state‑of‑the‑art methods on MovieLens1M, MovieLens 10M, and Netflix datasets, and adding more hidden layers can further improve the performance.