A Dual-view De-popularity Bias Recommendation Method Based on Uninteresting Item Mining
Popularity bias is a common problem in recommender systems,which causes popular items to be over-rec-ommended and non-popular items to be ignored.The existing research on popularity bias debiasing mainly improves the exposure of non-popular items by reordering the recommendation results or integrating regularization in the training pro-cess,but data sparseness and indistinguishability of users'preferences for popular items and non-popular items have be-come obstacles to further research.Aiming at this problem,a two-view popularity bias debiasing method is proposed based on uninteresting item mining,which mainly solves two problems:(1)Mitigate the impact of data sparsity and user selection bias on recommendation results by mining uninteresting items;(2)Dual-view learning is used to ensure that the exposure of non-popular items is improved while taking into account the preferences of different users for popular and non-popular items.In order to verify the effectiveness of the method,analysis and comparison are carried out on three public datasets.The experimental results show that the proposed method can effectively alleviate the popularity devi-ation while improving the non-accuracy performance.
NETL
NSTL
万方数据
维普
