Collaborative Filtering Recommendation Model based on Convolutional Denoising Auto Encoder
Huo Huan, Wei Zhang, Liang Liu, Yang Li · 2017
Sparse matrix problem and cold-start problem have been major challenges for recommendation systems. Many approaches take advantage of the combination of Denosing Auto Encoder (DAE) and Collaborative Filtering (CF) methods to address the above problems. However, most DAE-based algorithms adopt the bag-of-words model to process input texts, which loses the latent features of the context information. Therefore, this paper proposes a novel collaborative filtering recommendation model CDA-MF (Convolutional Denoising Auto Encoder-Matrix Factorization), which combines the Convolutional Neural Network (CNN) with the DAE and integrates into the Matrix Factorization algorithm so as to fully explore the latent context features. The experiments over real data verification show that CDA-MF model outperforms the state-of-the-art approaches and gracefully solves the cold-start problem on sparse matrices.