Collaborative filtering algorithm based on denoising auto-encoder and item embedding
Yudong Guo, Yongwang Tang · 2017
An collaborative filtering algorithm based on Denoising Auto-Encoder and item embedding (CDAWE) was proposed to solve the absent analysis of item co-occurrence relation and the cold start of model parameters of the information recommendation algorithm based on Denoising Auto-Encoder. In the proposed information recommendation algorithm, users are viewed as documents and items that users have rated are viewed as terms to form the training corpus firstly. Secondly, corpus are trained by the word embedding model, getting item embeddings that imply context information. Thirdly, the Denoising Auto-Encoder neural network is constructed by using all item embeddings as the initial weight and the model parameters are gained through training. Finally, user ratings are predicted by the model and the top-N recommendation is accomplished. Experimental results on the standard dataset demonstrated that the proposed algorithm has higher recommendation accuracy compared to current mainstream algorithms.