Abstract: The traditional recommendation algorithm generally uses the user project score matrix to learn the potential factors, understand the user’s personal preferences and make recommendations, but in practice, the score matrix is usually very sparse. Aiming at the shortcomings of traditional recommendation algorithm, a recommendation model based on noise reduction automatic encoder is proposed. First, two automatic encoders are used to train the potential factor matrix of the user and the project, and then the implied feature vector is input into a neural network to carry out the scoring prediction. Finally, the recommendation is made according to the new score matrix. The experimental results show that the proposed algorithm improves the recall rate of the recommended results and reduces the reconstruction error.

Key words: autoencoders, deep learning, neural network, collaborative filtering, denoising autoencoders