Research On Collaborative Filtering Recommendation Method Based On Adversarial Generative Network And Attention Mechanism

In recent years,with the rapid development of computer technology and the increasingly mature recommendation system,people often recommend products automatically by the system in online shopping.Since most recommendations are based on collaborative filtering recommendations,there is no more comprehensive,diversified,and in-depth consideration of user preferences.How to make more accurate and efficient recommendations for people's favorite has become a problem that researchers need to solve.This paper uses a collaborative filtering recommendation method that combines generative adversarial networks and attention mechanisms.The specific research is as follows:(1)This paper proposes a collaborative filtering model based on dual attention mechanism(CF-DA)mainly to solve the user's behavioral preferences and linear and nonlinear feature extraction.Since generating adversarial networks to make recommendations cannot solve the user's behavioral preferences,it can be solved using a two-layer attention mechanism.First,the user's information is obtained,and each user's preference for movie features is extracted as the input of the second layer in the process of the first attention layer.Then in the second attention layer,the user vector is combined with the user's viewing history and calculated,and the resulting value is used as the user's preferred feature value.Finally,the final weight value is obtained through a specific function using the preference eigenvalue and the linear eigenvalue.(2)This paper proposes a model that combines a generative adversarial network and an attention mechanism.In this paper,we first use user and movie information as input,and let the generator in the adversarial generation network generate an ndimensional vector.Next,this paper proposes a new method of fusion attention mechanism,which is to add a CF-DA model to the adversarial network model,and let the n-dimensional vector and the weight value obtained by the attention mechanism be integrated to calculate the result value.Put it in the recommendation list as the input of the discriminator.The discriminator will compare the input value with the real data.If the value of the recommendation list is close to the real data,the discriminator will discriminate it as true(real is 1)and make a recommendation;otherwise,discriminate If false(fake is 0),back propagation is performed.In the training process,each forward propagation will get the loss value of the output value and the true value.The smaller the loss value,the better the model.We will output the output of ACFGAN(the model of the combination of GANs and CF-DA)The value is used as an input to calculate the loss function.The generator and the discriminator will have their own corresponding loss function.The entire model uses the back propagation algorithm.The discriminator parameters are updated first,and finally the generator parameters need to be updated by sampling again.Noise data to optimize the model.(3)Experimental design,results and analysis.This article is based on the proposed method has better performance compared with other advanced methods.Finally,through a lot of experiments on the real data set,it is verified that the CF-DA model integrated into the ACFGAN model can effectively improve the recommendation accuracy.