Personalized Movie Recommendation System Based On Deep Learning And Object-aware

With development and popularization of the Internet,the amount of information generated by various applications is rapidly expanding.In the massive data,although users can enjoy rich and diverse information,the huge data scale makes it difficult to choose the content they need,resulting in the problem of information overload.The recommender system can greatly alleviate such problems.It can not only meet users’ demand for high-quality information,but also increase the attractiveness of enterprises to users and consolidate their position in the industry.In recent years,the emergence of deep learning has solved the problem of feature learning and high-level feature interaction in traditional recommendation systems,and fully exploited the highlevel feature interaction between different feature domains,thereby improving the prediction accuracy of the model.Therefore,it is widely used in recommendation system.Since then,with the advancement of computer technology and multi-size embedding technology,the neural network model based on multi-size embedding has been greatly improved in terms of the accuracy of prediction results,thus becoming one of research hot spots of current recommendation algorithms.Among them,ESPAN(Embedding Size Adjustment Policy Network),which dynamically adjusts the size of the embedding according to the frequency of users and items,has received extensive attention.In recommendation systems based on deep learning,embedding is a widely used important technique while the existing models based on multi-size embedding do not pay due attention to the user’s preference when constructing the embedding,which will lead to inaccurate prediction results.In this paper,the ESPAN model is deeply studied,and an object-aware neural network model ONN is proposed and implemented.The main research contents are as follows:(1)When building the embedding,many current methods focus more on the embedding size search and ignore the importance of user preference.In view of the above shortcomings,this paper pays attention to user preferences,fuses them with multi-size embedding and proposes an objectaware neural network ONN(Object-aware Neural Network).The experimental results show that the ONN model has a great improvement in accuracy and loss value compared with the baseline model,which proves the effectiveness and advancement of the ONN model.(2)The ONN model still needs to be improved on the feature cross problem of user and item embedding,so this problem is optimized by introducing feature cross technology,and an ONN＿FC(Object-aware Neural Network with Feature Cross)model based on enhanced feature cross is proposed.The model divides the multilayer perceptron and the feature cross layer into two upper and lower layers.The feature cross layer receives the input and performs the feature cross operation,and then connects the output with the embedding of the user and the item,and finally inputs it to the multilayer perceptron and obtains recommendation result.(3)Based on the proposed recommendation algorithm,this paper designs and develops a personalized movie recommendation system based on B/S architecture.From requirement analysis to system design,the functions of user presentation layer,back-end logic layer,data storage layer and recommendation layer are planned.OPN_FC is used as the model of the ranking module.Finally,the system is implemented using Django,Vue,Redis,Nginx,My SQL and Py Torch.