Research On Personalized Movie Recommendation Algorithm Based On Repurchase Behaviour

With the development of information technology,recommendation system has attracted more and more attention from researchers because of its convenience to users.However,existing recommendation algorithms have many problems in data feature extraction and feature vector coding.Aimed at above problems,this paper widely refer to the current mainstream recommendation algorithm,and analyze their advantages and disadvantages,respectively,using logistic regression is proposed a fusion of user preferences collaborative filtering algorithm,a deep learning neural network combined with Adaboost recommendation algorithm was proposed,which used to implement the recommendations of different types of data sets.In this paper,my primary working is following:(1)Aiming at the problem of insufficient utilization of hidden features existing in traditional collaborative filtering,this paper extracted the rating times of users for the same category items as features.The user similarity based on Pearson similarity was constructed by using the user's historical rating information,and cosine similarity about user attributes was constructed by using the user feature vector.Finally,the two similarity degrees were mixed.In addition,the average score of categories is used to reduce the sparsity of the scoring matrix used in Pearson similarity calculation.The effect of feature reduction index on user feature similarity is proposed.Dynamic similarity mixing based on Jekard is proposed to replace linear similarity mixing.Finally,logistic regression is used to predict the user's item preference tendency,and the score amplification index is used to modify the score predicted by the model.(2)For the data set with large amount of data,this paper chooses to use deep neural network for data fitting.However,while neural network has good fitting ability to big data set,it also has instability,such as local optimal solution.Adaboost can combine several weak learners in a certain way to form a strong learner with good performance and improve the accuracy of prediction.In view of the shortcomings of neural network,this paper combines deep neural network with Adaboost,and takes deep neural network as the base learner of Adaboost.It can play the advantages of Adaboost and avoid the disadvantages of unstable training of deep neural network.To solve the problem of high sparsity of item category feature vector,word vector is used to optimize the item category feature coding in this paper.A separate neural network is used to convert item category coding into dense vector,which is then input into Adaboost neural network,effectively reducing the sparsity of vector.In response to the gradient dispersion problem existing in the excitation function TANH,this paper uses the Leaky Relu function to optimize it,which can not only achieve the nonlinear requirement,but also achieve the slight update of the gradient.The experimental results are analyzed.The results show that the average absolute error of the collaborative filtering recommendation algorithm based on user repurchase behavior is about 0.705,which reduces by about 1.4%compared with the original collaborative filtering algorithm based on user.The error of Adaboost recommendation algorithm based on neural network is reduced to about 0.686,which is obviously better than other algorithms,and also verifies the superiority of the proposed recommendation algorithm.