Research And Implementation Of Product Recommendation Algorithm Based On Attention Mechanism

In recent years,with the advent of the era of big data,data mining has become a research hotspot of various experts and scholars.With the continuous rise of data mining,recommendation system,as an important branch of data mining technology,has also received more and more attention.Recommendation system can provide users with accurate recommendation services,especially in the field of e-commerce,all kinds of recommendation algorithms play an increasingly important role.Among all kinds of recommendation algorithms,traditional recommendation algorithm and depth model-based algorithm have been widely studied.The traditional recommendation algorithm and depth model-based recommendation algorithm have been proved by numerous studies to improve the accuracy of recommendation.Some researchers combine the two methods to form complementary advantages,which further improves the accuracy of recommendation.However,there are still data sparsity and cold start problems in these hybrid algorithms.Moreover,for the personalized recommendation system based on user interaction sequence,the existing recommendation algorithm based on cyclic neural network can not fully mine the hidden user interest in user interaction sequence.Therefore,in order to solve the above problems and improve the accuracy of personalized recommendation algorithm in the field of e-commerce,we do the following research.The specific work and main innovations of this paper are as follows:1.Firstly,this paper introduces the background and significance of this research,as well as the research status of traditional recommendation algorithm and neural network based recommendation algorithm at home and abroad.Then,the related technologies in the field of recommendation system are described,including the model of recurrent neural network and attention mechanism.Finally,this paper designs a personalized product recommendation model based on attention mechanism.2.This paper proposes the application of deep learning model and attention mechanism model in DattRec recommendation algorithm.Based on the existing work,most of the personalized product recommendation algorithms simply regard the user's behavior as interest directly,while the implicit interest of the user in the interaction behavior is difficult to express through the explicit behavior,and most of the recommendation algorithms do not consider that the user's interest is evolving.To solve this problem,this paper proposes a hybrid personalized product recommendation model based on encoder-decoder structure.On the encoder side: the model models the internal relationship between the historical interactive products of usersthrough the multi-layer self-attention network,and then inputs the output of the multi-layer self-attention network into the GRU network with attention score to model the evolution process of users' interest.In the decoder side: This paper uses bilinear interpolation matching function to match candidate and user interest.3.Compare the model proposed in this paper with eight typical recommendation models on two datasets.The experiment shows that the model proposed in this paper achieves the most advanced recommendation accuracy.By adjusting the parameters,the influence of the parameters on the experimental results is explored.4.Finally,in order to verify the effect of the model proposed in this paper in practice,a prototype system of commodity recommendation based on flack framework is developed.With the help of this specific implementation,the recommendation effect of the model proposed in this paper in practical application is fully verified and demonstrated.