Research Of Recommendation Algorithm Based On Deep Latent Features Analyzing

With the rapid development of Internet industry,the amount of information on the Internet is growing exponentially.How to choose the information users want from all kinds of Internet information becomes the main task of the recommendation system.Recommendation system achieve the purpose of predicting user behavior next by analyzing the user's historical data effectively,and become one of the main means to solve the information overload problem at present.However,the traditional recommendation algorithm based on collaborative filtering was proposed for users-item rating data modeling analysis,generally uses the linear model,which lead to the bottleneck for improving the accuracy of the recommender.The deep learning technology with strong ability of characterization has achieved great success in the image recognition and natural language processing,so the recommendation algorithm based on deep learning is widely studied in recent years.To address the issue of weak data representation ability of traditional collaborative filtering algorithm,this paper,by combining relevant theories and methods of the deep learning,studies the recommendation algorithm based on deep latent features analyzing,aiming to improve the accuracy of the recommendation algorithm.The specific work is as follows(1)Studying the present situation of recommendation algorithm widely,the recommendation algorithm based on deep learning has the advantage in dealing with multi-source data and data representation.In the actual recommendation system,information data of users and items is varied,such as image information of the item Recommendation algorithm based on deep learning can handle these data by using a variety of network model,and extract more representative features compared with the traditional collaborative filtering algorithm.In addition,the recommendation algorithm based on deep learning can carry on the nonlinear characteristics of users and items,establish a nonlinear model,and improve the ability of generalization.(2)A collaborative filtering algorithm based on latent feature analysis is proposed.The latent features discussed in this paper refer to the features extracted by the neural networks.This method analyze data from the angle of users and items,establish two different recommendation models based on the user and item by using the autoencoders respectively.These models analyze the latent features extracted from the feature vector of users and items and introduce a variety of regularization methods and denoising techniques,to alleviate overfitting problem,enhance the robustness of these models.Through multiple sets of experiments on multiple data sets,the correctness of the method is verified(3)An integrated recommendation algorithm based on the deep latent feature fusion is proposed.This method fuse the latent features of users and items by using multilayer perceptron,and extract the interaction features between the users and items.In addition,to address optimization problem of deep neural networks,various solutions are proposed,such as batch normalization.And pre-training method of the model are used to improve forecasting precision.The experimental results show that this integrated recommendation algorithm based on the deep latent feature fusion has obvious improvement in precision compared with the existing recommendation algorithms.