Research Of Hybrid Recommender System Model Based On Multi-Dimensional Features

Nowadays,with the widely using of Internet and rapidly development of Ecommerce,all kinds of information grow explosively,and users face a large number of information which is irrelevant to the users,which made difficulties for users to obtain what they need.To solve this problem,more and more recommender systems which provide personalized services come into being.The personalized recommender system aims to eliminate a lot of irrelevant information for users and provide personalized things to users.Different recommendation algorithms are usually used for all kinds of recommender systems with different scenarios,and the most commonly used ones are content-based recommendation and collaborative filtering recommendation.Contentbased recommendation and collaborative filtering recommendation have a lot of problems,like the cold-start problem,the quality of recommendation being seriously depressed by sparse ratings of users,bad system expansibility,low level of efficiency,etc.In order to solve these problems and improve the quality of recommendation,many researchers propose a lot of ideas and solutions in different recommender scenarios.This thesis presents the research of a hybrid recommendation method based on content and collaborative filtering,and the solution of how to solve the problem of data sparsity,recommendation efficiency and so on.In addition,this thesis argues that the user's visual information for the study of recommender systems has very positive effect,so in this paper,the user's visual information was used to explore the possibility of applying it in recommender systems.The main content and contributions of the thesis are:(1)Most recommender systems are based on user static profile data or dynamic interactive data to calculate the user interest model.This thesis attempts to use data from multiple dimensions to build the user model.In addition,this thesis uses image visual features,which are often ignored by researchers,to find users' potential preferences.(2)Based on the advantages of content-based approaches,this thesis proposes a calculation method of users' interest measure of multi-dimensional features,which solves the data sparsity problem and recommendation efficiency problem and so on.Therefore,this recommendation method satisfies both extremely sparse data set and larger data set.(3)This thesis combines the content-based user modeling method and collaborative filtering recommendation method to propose a hybrid recommendation model based on multi-dimensional features,which makes our recommendation model suitable for a variety of different rating-based recommendation scenarios.Besides,the adaptability and scalability of the recommendation model are improved.(4)Plentiful of experiments are performed on the real world MovieLens datasets to evaluate the proposed model.The experimental results show that the proposed approach can improve the recommendation accuracy on the sparse datasets and has higher efficiency on the large scale datasets compared to the baseline methods.