Research On Collaborative Filtering Algorithms Combining Attributes And Implicit Social Information

With the rapid growth of the Internet,it is very difficult for users to find useful information and what they are interested in the vast amount of data.The recommendation system came into being.Due to its wide application prospect and research value,the proposed algorithm has been closely watched.In recent years,the rise of various online recommendation services for recommendation al-gorithm proposed many new challenges to recommendation algorithm,such as cold start problem.Traditional recommendation algorithm based on historical records is unable to produce accurate recommendations for newly registered users or newly released items.Expansibility,when the number of users and items grows rapidly,the memory based method cannot achieve good online prediction,so it is not scalable.Recommendation precision,recommendation accuracy affects user's experience,good recommendation accuracy can increase user's stickiness.Aiming at these challenges,the innovation points of this paper can be divided into the following points:· Propose an improved algorithm for user cold start problem——an improved collaborative filtering algorithm based on coupled user attribute.Based on memory based collaborative filtering framework,the improved algorithm capture the real relationship between category type data by improved coupled similarity.This improved algorithm build a similarity model on user attributes.The improved coupled similarity can better capture the similarity rela-tion between users,so the improved collaborative filtering algorithm based on coupled user attribute can also provide a collection of items for new users and achieve better recommen-dation accuracy in the cold start-up stage of the user client.· Propose two improved algorithm that have extensibility and can deal with the cold start-up of the item client——an improved collaborative filtering algorithm based on the coupled average content attribute and an improved collaborative filtering algorithm based on the coupled individual content attribute.The improved algorithm introduces the regulariza-tion items of the average content attribute information and the individual content attribute information respectively in the matrix decomposition framework to constrain matrix de-composition process.Thus,the implicit eigenvectors of objects whose content attributes are similar are similarly learned.Compare to the improved collaborative filtering algorithm based on the coupled average content attribute,the improved collaborative filtering algo-rithm based on the coupled individual content attribute builds trust propagation and avoids the disadvantages of averting average neglect of individual diversity,thus the improved collaborative filtering algorithm based on the coupled individual content attribute further improves the recommendation performance.Due to the introduction of content information and matrix decomposition framework,the improved collaborative filtering algorithm based on coupled content attribute have extensibility,and it can provide the items to users in the cold start-up stage of the item client.Besides,the introduction of the additional information can also improve the recommendation accuracy.· Propose an improved algorithm for improving recommendation accuracy-an improved collaborative filtering algorithm based on content and implicit social information(CBNMF algorithm).The improved algorithm introduces the implicit social information regulariza-tion in the matrix decomposition framework that coupled content information to constrain matrix decomposition process.Ratings behavior of users is equal to users' performance in social communication.Therefore,user trust in implicit social information is replaced by the user similarity profile based on the score,and the more similar users are,the more the user trusts her.The idea comes from the trust spread of social network users,that is,users will consider the opinions of a group of friends around them when choosing an item,and the more they trust their friends,the higher the adoption rates will be.The CBNMF algorithm build the trust propagation.If user A trusts B and B trusts C,the user implicit eigenvectors of A and B,B and C,A and C learned from the matrix decomposition will also be as similar as possible.Experiments show that the proposed CBNMF algorithm in this chapter can ac-curately learn the characteristics of users and items vector,the accuracy is superior to other benchmark algorithms,and it can effectively improve the user's recommendation accuracy when there are more user scores.Of course,it's even better if you have explicit user social data,because it can further improve user client cold startup problems.