Hybrid Recommendation Algorithm Based on Adversarial Learning-to-rank
The recommendation system can help users filter the massive amount of information.Every single recommendation algo-rithm has some defects.Mixed recommendation based on deep learning can effectively alleviate the problem of sparse data in tradi-tional recommendation algorithms by incorporating auxiliary information and often achieve better results.In most current studies,specific auxiliary information is used for different algorithms,but there is no unified hybrid recommendation framework.This paper proposed a hybrid recommendation algorithm based on adversarial learning-to-rank:MRecGAN.The idea of learning-to-rank was used to integrate multiple basic recommendation algorithms,built unified auxiliary data,and digged the deep relationship among fea-tures.It used generative adversarial networks to learn the sorting function,improved the performance of one discriminator and two generators,and obtained the recommendation sequence.Finally,the real Movielens dataset combined with auxiliary data was used for testing.The experimental results show that the model integrates the advantages of the basic models better.NDCG and other in-dexes improve significantly,MRR improves by 32.05%compared to CML.
国家科技期刊平台
NSTL
万方数据
