Collaborative Convolution AutoEncoder for Recommendation Systems.

Against traditional collaborative filtering algorithm often face the problems of high sparsity and low recommendation accuracy, we propose collaborative convolution autoencoder algorithm CCAE, which replaces the matrix decomposition training method in the collaborative filtering algorithm with the convolution autoencoder method. First, the input data drop sampling through the convolution layer and the downsampling layer to learn its efficient compression characteristics, then the data reconstruct through the deconvolution layer and the upsampling layer, and calculate the score ranking for recommendation. Experimental results show that the algorithm CCAE achieves lower RMSE than those based on autoencoder AE, stack denoising autoencoder SDAE and collaborative filtering matrix decomposition MF on movie-lens. Therefore, CCAE algorithm can effectively solve data sparsity and improve recommendation accuracy.