CF-SSDAE: Symmetric SDAE for Collaborative Filtering Algorithm

摘要

Collaborative Deep Learning (CDL) uses the stacked denoising autoencoder to encode information matrix of items, which reduces the dimension of the matrix, but the method does not consider the user information. In order to get the user information, a new acquisition method is proposed: the user information matrix is obtained from the item information matrix and the rating matrix, so that the user information matrix and the item information matrix are in the same space. And a symmetric SDAE for collaborative filtering algorithm(CF-SSDAE) is proposed, which uses the two stacked denoising autoencoders to simultaneously train the user information matrix and the item information matrix to obtain the feature matrix of users and items. In order to compare with the state-of-the-art algorithms, experiments show that our algorithm has higher accuracy on the CiteULike dataset.