Research And Application Of Collaborative Filtering Recommendation Algorithm Based On User-similarity In Social Networks

Data overload problem is a constraint issue in developing of the Internet.Personalized recommendation technology effectively alleviated this problem,but with the Internet system complexity,data content diversification,personalized recommendation technology applications exposed some other problems.Such as data cold start,sparsity of user data,recommendation system scalability issues.Collaborative filtering recommendation algorithm generates recommendations to users which they are interested in through according the users' preferences.It provides recommendations to users from massive data resources.Collaborative filtering recommendation algorithm has been widely used in e-commerce.However,when this type of algorithm is applied to social networks,the focus of traditional evaluation index and similarity calculation was changed.It came to appear recommendation algorithm efficiency is low,recommended accuracy is also low,resulting user satisfaction is low online dating in a social network.In response to solve these problems,in this paper,we designed user-similarity based collaborative filtering recommendation algorithm,and verify the effectiveness of the algorithm in a simulation environment.Main tasks of paper are as follows:(1)Redefine the user similarity and calculation methods.User similarity is made of two parts linear fit with them:? User attributes similarity contains user information and attributes.It divides user attributes into numeric attributes and name numeric attributes.? User interaction similarity.Session between the user based on information to identify the sender of similarity between users and similar recipients expense to measure user interaction similarity values.? Both similarity are given different weights,linear fit similarity between them.Generate recommendations for user by order users using the Top-N sorting users similarity.(2)Designed collaborative filtering recommendation algorithm based on user-similarity and discuss basis on algorithm design.Generate recommendations for users by calculate the similarity with Top-N algorithm sorting candidate set.Analyse the complexity of algorithm and compare complexity with other algorithms.(3)Built experiment environment.Simulation environment utilizes Apache Mahout open-source platform.We use social networks real historical data.Then,collect experimental data and analysis data.Evaluate algorithm on accuracy,user response rates,coverage.Made experiments on three different algorithms:collaborative filtering recommendation algorithm,interactive based recommendation algorithm,recommendation algorithm based on user similarity.Three algorithms were evaluated.Simulate real environment under social networks offline to really maximize close to the real situation.The experimental results showed:Compared with others,the complexity of user-based similarity filtering algorithm is in the same order of magnitude with them.Without increasing the additional cost in the case,based on user similarity coordinate filtering recommendation algorithm in social networks got a higher quality results than the other two algorithms,and got a higher user satisfaction.