Research On Commodity Recommendation Algorithm Based On Deep Learning And Collaborative Filtering

The rapid development of Internet and the growing maturity of data mining technology have promoted the information construction of social life in all aspects,as well as affected our daily life profoundly.Meanwhile,various information resources have been increasing explosively which cannot be effectively integrated into the content that people need,resulting in the phenomenon of information overload.In the field of e-commerce,the recommender system filters products by analyzing and mining the interests of users.It recommends suitable products for users to meet their personal demands,and improve the economic benefits of businesses.However,traditional collaborative filtering methods are difficult to mine the correlation between users and items,and the sparsity of the rating matrix limits the ability to predict of recommender system.As review text can directly reflect user preferences and model users and items more comprehensively,more and more attention has been paid to review data,which can be used as a supplement to rating data to optimize the recommendation effect.The current methods have some shortcomings in the feature extraction of review data and the fusion of user/product features.Therefore,this paper proposes a deep hybrid recommendation model with gating mechanism.Firstly,in order to learn the contextual information and personalized representation in the review data,a method of bidirectional gated recurrent unit network combined with attention mechanism is proposed to extract review features For the learning of rating features,multi-layer perceptron is used to model rating representation based on embedding layer.Then a gated fusion mechanism is proposed to learn the importance weights between different dimensions,which can adaptively fuse the two features.Finally,the interaction between features is enhanced through the feature interaction layer,and the correlation information is retained,then the factorization machine is used for rating prediction.We compared the performance with multiple baseline models on the Amazon product dataset and Yelp review dataset.The experiment results show that our method not only has lower prediction error,but also has better generalization.It improves the recommendation performance to a certain extent and has good interpretability for the recommendation results.