Research On Personalized Recommendation Algorithm Based On Deep Learning Theory

With the rapid development of mobile cloud computing,big data,artificial intelligence,and 5G technologies,China has entered a new era of Industry 4.0.At present,there are more and more types of information service platforms provided by small and micro Internet companies through mobile phones and Internet clients,such as Kuaishou,Douyin,Weibo,and We Chat circle of friends.The popularity of these software products has caused an explosive growth in data scale.In the face of these data,how to quickly and accurately mine useful information is very important,and these platforms have innovatively proposed their own core recommendation algorithms to allow users to passively accept the information they are interested in.Under this development background,the recommendation algorithm under the platform has become a hot research topic.This paper studies the recommendation algorithms by studying and using deep learning technologies such as convolutional neural networks and autoencoders,and proposes two personalized recommendation algorithms based on deep neural networks.1)Video personalized recommendation algorithm based on deep auto encoder with text convolution.The algorithm combines a convolutional neural network and a deep autoencoder network,uses the auxiliary information of the video dataset(such as user information and item information)to deeply dig potential information,and introduces a new deep learning-based recommendation model for iterative training Potential links between video user,item,and ratings,Which in turn predicts that users will recommend ratings for unknown items.Compared with the traditional recommendation algorithm,this algorithm not only improves the recommendation accuracy,but also solves the problems of sparsity and cold start in the recommendation algorithm.2)Video personalized recommendation algorithm based on scalable multi-channel fusion strategy auto-encoder model.The algorithm combines the self-encoding network and the image convolution multi-channel idea,and uses the user's historical interactive scoring preference matrix to mine the potential characteristics of users and items.Furthermore,based on the iterative training of the established model,the connection between the user,the item,and the rating is obtained to predict the user's recommendation for the unknown item's rating.Compared with the traditional recommendation algorithm,the recommendation accuracy of this algorithm has not only greatly improved,but the model is also very flexible and versatile.In the end,the proposed algorithms have been experimented to test the performance of the proposed algorithms.Compared with traditional recommendation algorithms,it has improved on different evaluation indicators.