Personalized Movie Recommendation System Based On Deep Learning And Behavior Sequence

In recent years,the rapid development of the Internet has led to the exponential growth of the amount of network information.Searching for information in the vast information flow has always been a headache for users.It was not until the emergence of intelligent recommendation systems that this problem was greatly alleviated.It not only solves the problem of users’ difficulty in searching for information,but also brings huge benefits to enterprises themselves.In this background,the recommendation system in the field of movies has also followed.Because of the powerful feature learning ability of deep learning,it has become the most popular recommendation algorithm in movie recommendation system.Among them,extracting user behavior sequence information through deep learning for recommendation is the focus of many researchers today.Existing deep learning recommendation algorithms are not sufficient to extract the dependent information in the user behavior sequence,and there is still a lot of room for improvement.At the same time,how to accurately capture the user’s dynamically changing interest from the behavior sequence is also a key issue.Based on the above problems,this paper proposes two recommendation algorithm models and embeds them into the personalized movie recommendation system.The main work of the paper is as follows:(1)Based on the attention network,this paper proposes a recall model UEIIN(User Explicit and Implicit Interest Network)which combines explicit preference and implicit interest.The model first uses the multi-head self-attention module(a module that adds optimization measures to the multi-head self-attention mechanism)to extract the implicit interest in the user’s behavior sequence,and at the same time uses explicit feedback information to extract user’s explicit preference.In the actual scene,the user’s interest is constantly changing dynamically,so we use the basic attention mechanism to learn the dynamic relationship between the user’s explicit preference and implicit interest,and input the final learned explicit preference and implicit interest into the deep neural network to learn deeper information.Finally,the experimental results on datasets with different sparsity show that the proposed model achieves better performance in Recall,NDCG(Normalized Discounted Cumulative Gain)and Hit Rate.(2)This paper proposes a ranking model DBIN(Deep Behavioral Interactive Network)combines factorization machine and multi-head self-attention module.The model divides user behavior interaction into first-order,second-order,and high-order interactions.The model retains the original sequence behavior as the first-order interactive behavior feature,which will make the model have better memory to a certain extent;the second-order interactive behavior is obtained by factorization machine,which is the second-order cross behavior feature representation in learning behavior sequence;the multi-head self-attention module with the auxiliary loss function is used to capture the feature representation of high-order interactive behavior,the auxiliary loss function can significantly improve the accuracy of the feature representation.Finally,the basic attention mechanism is used to learn the influence factors of different types of interactive behaviors on candidate items,and use the multilayer perceptron is used to predict the user’s click probability.The model uses AUC,a commonly used test index in the field of click through rate estimation,to test the ranking performance.At the same time,it uses Rela Impr to measure the relative improvement of the model.The experimental results show that the DBIN model is used in three different public datasets and one private dataset(Douban movie)compared with the current popular similar models,it has achieved a good improvement.(3)Based on the recommendation algorithm proposed in this paper,a personalized movie recommendation system is designed and developed.Firstly,the system uses the model UEIIN that combines the user’s explicit and implicit interests to recall the N movies that users are most interested in from the complete movie library,and then uses the ranking model DBIN based on user behavior interaction is used to rank the N movies from large to small according to the click through rate.Finally,the sorted movie list is output to the user.This architecture significantly improves the accuracy of personalized movie recommendation,which has good practical significance.