SDAE-LFM: A Latent Factor Model for Recommendation Based on Stack Denoising AutoEncoder

Abstract Recommendation methods usually associated with data sparsity. The traditional recommendation methods take the users’ rating information as the recommendation basis, which ignore the latent features that can be taking into consideration to model for better recommendations. In order to deal with these problems, we proposed a latent factor model recommendation algorithm based on stack denoising autoencoder (SDAE-LFM), applying Deep Learning technology for latent feature representation learning. A stack denoising autoencoder is applied to extracting feature about item from the label information. Then we factorize the item feature information to perform matrix decomposition training. Finally, we predict the result by the user-item preference matrix. Experimental results on these datasets demonstrate that the proposed recommendation method has better performance.