A Study Of Recommendation Based On Collaborative Filtering

With the rapid development of the Internet,the mankind has entered the era of network information,which provides people with more and more rich services and experience.We can break through the restrictions of time and space,do what we want to do on the Internet,such as chatting,watching news,watching videos,playing games,shopping,learning and so on.However,the rapid development of the Internet also makes a large variety of information grow rapidly which filled with a large number of useless information for users and it is difficult to find useful information from a large number of the information.It decrease the efficient of information useage and resulted in information overload.Personalized recommendation system is an effective solution to solve this problem.It recommends things to users which they like based on user's needs and interests and make users have a better experience on the internet.The core of the personalized recommendation system is the recommendation algorithm.And collaborative filtering is the most widely known and widely used recommendation algorithm.Its basic thought is similar to go to the supermarket to buy things,the probability of purchasing the things of most people like will be very high for us;Or to one thing we like and when we see an other thing which is similar to the thing and other people like it,then the probability we buy it will be very high.Collaborative filtering relies on the user-item rating matrix.There are still many problems in collaborative filtering algorithm,such as cold start problem and data sparsity problem.Its accuracy of recommendation is still needed to be improved.For cold start problem,this paper proposes an improved method based on classification and feature similarity.For the problem of data sparsity,this paper proposes a collaborative filtering algorithm based on improved multi-segmented PCC and time characteristic in order to solve the problem.The improved method based on classification and feature similarity uses the content information of the new user to classify users,and then uses feature information to calculation similarity to obtain the nearest neighbors of the current user,and then predicts the score of the current item for the new user.Finally,the method recommends several items which have highest scores to the current user.The method makes up for the problem that new users or new items have no rating datas,and solves the cold start problem to a certain extent.The collaborative filtering algorithm based on improved multi-segmented PCC and ime characteristic makes the PCC divided into multiple levels according to some conditions,considers the number of co-items and the PCC threshold as partition conditions.Then adding a positive value to the PCC calculation results instead of adding a weighted number of co-items to the PCC results.In the prediction stage,a time weight is added.Finally,the proposed method is compared with collaborative filtering which use other similarity calculation method through experiments.The experimental results show that the proposed method is better than other traditional methods.