Based On Collaborative Filtering Personalized Recommendation Research

With the rapid development of the computer and the Internet,users are exposed to more and more information and services.However,it also brings users the problem of "information overload".Personalized recommendation has become the most effective way to solve this problem,and has attracted more and more researchers' attention.Personalized recommendation can mine the interests of the user from the massive data to recommend possible valuable information.Collaborative filtering recommendation system is currently the most successful and one of the most widely used technology.However,the existing collaborative filtering algorithms are several major problems: data sparsity,cold start,extendibility,precision and ignoring the information of users and items.This paper concentrates on the collaborative filtering technology,alleviates the problem of data sparsity,cold start,extendibility and ignoring the problem of user and items information from two aspects: the safe high-confident semi-supervised algorithm and dimension reduction based on neural network,clustering.The main works of this paper is:1)Researches the existing semi-supervised algorithm and proposes a collaborative filtering recommendation algorithm based safe and high confident semi-supervised method(Safe-Simi-supervised-based Collaborative Filtering,SSCF).This algorithm uses safe and high confident Semi-supervised method S4 VM to predict the rating of unrated items.Through predicting the rating of unrated items,the sparsity of data has been alleviated and the accuracy of searching nearest neighbor item has been improved simultaneously.2)Researches the existing algorithm and framework of the neural network model and proposes collaborative filtering algorithm based on cluster based on dimension reduction of autoencoder neural network model(AutoEncoder dimension reduction Cluster based Collaborative Filtering,AECCF).This algorithm first uses autoencoder neural network model to train user feature,which can represent user in low-dimensional and expressive feature,then uses k-means cluster user,finally,uses based on user collaborative filtering for every cluster.So,it improves the prediction accuracy and improves the scalability of model.3)Designs and implements movie recommendation system SSAE-film based on SSCF and AECCF algorithms.