Research on Recommendation Algorithms Based on Asymmetric Attention Mechanism Models.

Although models such as matrix factorization and graph neural networks have achieved good performance in rating prediction, they still face some problems, such as cold start and data sparsity. In this paper, we propose a model called Asymmetrical Local-Global Attention Model(ALGM) to address these issues by combining user comments and rating information. Firstly, to alleviate the data sparsity problem faced by previous recommendation algorithms that only use rating information, we introduce additional user comment information. Secondly, in order to extract valuable information from comments more effectively, we utilize dynamic word embedding tools for modeling. Finally, to accurately depict user and item representations, we adopt attention mechanisms and an asymmetric approach to model user and item features. The proposed algorithm in this paper is evaluated on datasets such as Amazon, using mean squared error (MSE) as the evaluation metric. Experimental results show that the ALGM algorithm proposed in this paper outperforms the selected baseline models by 2% in performance.