Research And Implementation Of Hybrid Recommendation Based On User Features

The breakthrough of Internet related technology brings great convenience for information transmission and real-time communication.At the same time,the increasing amount of data has become an obstacle to retrieve useful information in an effective time,causing the problem of "Information Overload".Recommendation system has gradually become a hot technology in the new network era by virtue of its powerful data filtering ability and the intelligence that can provide personalized information push service for users.However,there are still defects such as Cold Start and Concept Drift in recommendation system,which are new challenges to be solved in recommendation field.In this thesis,based on the features of users,the problems in the recommendation algorithm are optimized.By studying the advantages and disadvantages of the existing recommendation system,different solutions are proposed for the above problems.The innovation and contribution of this thesis are as follows:Firstly,this thesis analyzes the user interest drift in the process of recommendation.The user interest state shifts with the change of implicit conditions in the external environment.It is difficult to capture the change of external conditions,but the changed user state can be reflected by user behavior.Therefore,through the analysis of user's dynamic behavior,the concept of time characteristics is added to the user's behavior record of the project,and the utbcf algorithm based on user rating behavior time series is proposed to improve and optimize the time weight formula.The validity of the algorithm is verified on the MovieLens-100 k data set.The experimental results show that the optimized algorithm has better recommendation quality than the original algorithm.Secondly,through the analysis of the user's static portrait,the thesis proposes the recommendation USFBF algorithm based on the user's static characteristics and improve the user similarity calculation method.In the period of new users' transition,it can be used as a substitute for recommendation strategy to solve the Cold Start problem.In the case of less user rating,users are clustered by using their static feature,and then the items are clustered by item labels to establish the relationship between user groups and item groups for matching and recommendation.It is proved that the algorithm is a good recommended replacement strategy in response to the transition period of new users starting on the MovieLens-100 k dataset.Combining the User Static Feature Based Filter recommendation method with the User Timing Based Filter recommendation method,this thesis proposes a Hybrid Recommendation Based On User Features algorithm for dynamic switching of the number of different users' ratings recommendation method,which comprehensively solve the limitations of a single recommendation strategy in response to different user states.Finally,this thesis implements a web-based movie recommendation platform application,which uses React Ant-Designed to build the front-end page,and users interact with the platform online through the browser page.The background is mainly implemented by Flask framework,and the database is implemented by MySQL.The HRBUF recommendation algorithm introduced in this thesis is applied to the recommendation module of the platform.It can help users to retrieve and recommend movies of interest,and provide display of movie related information...