A recommendation algorithm for collaborative denoising auto-encoders based on user preference diffusion
Xiu Wang, Xuejun Liu, Xinyan Xu · 2017
The main challenge of collaborative filtering is to use a relatively less effective score to get an accurate prediction. It is an effective method to incorporate user and item information with collaborative filtering. In this paper, we present a new method. Firstly, the heterogeneous information network is used to combine more information of users and items. Under the corresponding user interest semantic assumptions, the observed user implicit feedback extends the user's preference items along different meta-paths. And then combined with the Collaborative Denoising Auto-Encoder (CDAE) for top-N recommendation. Denoising Auto-Encoders learns latent representation of corrupted user-item preferences that can best reconstruct the full input. Experiments show that the proposed algorithm can effectively alleviate the sparse data and improve the performance in terms of recommendation accuracy.