| With the development of Internet technology,recommendation systems are used in more and more scenarios to alleviate the problem of user information overload in the era of information explosion.A good recommendation system can greatly reduce the manpower and resources that users spend in the process of finding the required information.Time,improve user experience and create profits for the enterprise.Therefore,how to improve the performance of the recommendation system and quickly filter out the user's required information from a large amount of information has become a hot research topic nowadays.After years of development,the traditional simple recommendation model can no longer satisfy users' increasingly rich information.More and more researches try to use deep learning in the field of recommendation systems to realize the intelligence of recommendation systems.This article focuses on the research and application of the related algorithms of the recommendation system,and tries to combine the gradient boosting decision tree GBDT(Gradient Boosting Decision Tree)and the deep learning model WDL(Wide and Deep Learning)to realize a combined model that can realize automatic feature intersection GWDL,and the introduction of Stacked Denoised Autoencoder(SDAE)technology,by providing a high-dimensional dense feature vector matrix for the deep learning part of the model to facilitate denoising,to achieve the integration of SDAE and GWDL models,based on the above foundation,this article The hybrid recommendation algorithm SDAE-GWDL is proposed and established,and it is applied to personalized movie recommendation websites to recommend movies that may be of interest to website users and improve user experience.In order to achieve the above goals,the main work of this paper is as follows:1.Research the feature engineering part of the WDL model,and propose to combine the GBDT decision tree with the WDL model to solve the problem of manual feature intersection in the deep learning model WDL and improve the ability of the model to mining high-level features of users.2.Research the output layer of the model,and improve the design of the input layer of the model with reference to the idea of the attention mechanism.3.Research on the stack denoising autoencoder.Aiming at the problem of sparse feature data in the actual recommendation scene,it is proposed to combine SDAE with the above GWDL model to improve the accuracy and to improve the model Cold start performance.This paper conducts model performance test experiments on three commonly used data sets for recommended model training(Movielens data set,Lastfm data set and Jester data set),using Precision,AUC(Area Under the Curve),and Recall as the evaluation indicators of the experiment.The performance of the new model is evaluated by the average of the above evaluation indicators on the three data sets.The experimental results show that the Precision,AUC,and Recall of the basic model are 0.7397,0.8401,and 0.5591,respectively,and the index values of the improved model are 0.7626,0.8777 and 0.5934,the three indicators have improved to varying degrees,verifying the performance improvement of the algorithm model in recommendation.In addition,this paper also evaluates the cold start performance of the model with the evaluation index mAP(mean average precision),and the results show that the mAP of the improved model is increased by an average of 2.45%on the basis of the original model.The SDAE-GWDL algorithm model researched and proposed in this paper can improve the accuracy of the recommendation results and improve the cold start performance of the recommendation system,and can be applied to the actual recommendation system.Based on the above research on the recommendation model,this article also designs and implements a personalized movie recommendation system based on the Django platform.This system can achieve efficient personalized movie recommendation functions,improve the accuracy and cold start performance of the recommendation results. |
PDF Full Text Request