Research And Implementation Of Recommendation Model Based On Multi-feature Fusion Analysis

In recent years,the rapid development of the Internet has caused the explosive growth of information in online systems,and the demand of users for personalized information has become more and more intense.Recommendation system has become an effective tool for massive information filtering and accurate presentation.The application of recommendation systems in e-commerce websites and other online content consumption systems is also becoming more widespread.Common recommendation system methods are collaborative filtering algorithms and improved algorithms based on matrix factorization.But most of the methods are based on rating prediction mode,ignoring the temporal context and recurrent pattern of user behavior,without simultaneously considering the multiple features of user behavior under different situations.At the same time,the matrix sparse problem and scalability problem in recommendation systems need further improvement.This paper proposes a recommendation model based on multi-feature fusion analysis,which aims to comprehensively consider the temporal context and recurrent pattern of user behavior.Considering several explicit features,this new recommendation model also fuses implicit features into the analysis to improve the matrix sparse problem and scalability in the recommendation system research;thereby,further improving the accuracy of the recommendation system.Online systems generate large-scale user behavior logs every day,such as activity records in shopping,social,music,and video softwares,and we collectively call this kind of behavior user online content consumption behavior records.In online content consumption systems,user behaviors are divided into user long-term behavior based on time series,recent short-term behavior caused by user habit change,and user behavior under popular elements caused by popular factors at that time.Therefore,we can get the long-term user activity sequence and the short-term user activity sequence that are arranged by time.Besides,by considering the popular elements,we can filter the user's consumption content to get the user activity sequence under the influence of the popular elements at that time.Based on above three activity sequences in different recurrent patterns and temporal contexts,we propose a new,less error,and more effective recommendation model,which nests the recurrent neural network and the back propagation neural network in the dynamic recommendation system.Learning different types of user behaviors',features in the online content consumption system can more accurately predict the user's online consumption behavior at the next moment.Then the probabilities of users'consuming different items are converted into ratings to construct ratings matrix.The ratings matrix is used to fuse implicit features of users by Latent Factor Model to improve the traditional collaborative filtering algorithm,solving matrix sparseness and scalability problem in the recommendation system;thereby obtaining a more accurate recommendation model.At the last,a recommendation model based on multi-feature fusion analysis is designed in this study.The validity and accuracy of the recommendation model are verified in the real-world datasets,which has great practical significance.