To solve the sparseness problem of user-item interaction data in the target domain,a novel cross-domain recommendation model is presented,which can transfer the knowledges learned from the auxiliary domain into the target domain.The domain special features of the evaluation scores and comment texts can be extracted based on feedforward neural network.By using the adversarial model obtained by fusing the neural network based feature extractor and the vector embedding based domain discriminator,the common features of the evaluation scores and comment texts can be extracted.The domain special and common features are fused by using the multiple attention mechanisms,and the user-item interest can be solved.Experimental results show that for the Amazon dataset,the proposed model can achieve a better performance in terms of two classical recommendation evaluation indicators.
维普
万方数据