Research On Product Recommendation Method Based On Aggregative Statistics

Nowadays the users’ private information is disclosed for the purpose of privacy protect on many social media sites and e-commerce sites.For example,we know neither whether a user has browsed an article nor whether a user likes an article.The only thing that the sites make public is the aggregative statistics of an article or an item,which represents the total number of the users who like the article or the item.As a result,for the social sales or the shops who couldn’t achieve the sites’ internal information,the original recommendation methods,whether the methods based on content making use of the information of the items the user has bought or the collaborative filtering methods which take the user’ previous behaviors into use to calculate the similarity between users or items,are all out of application.To solve this problem,we study the product recommendation methods based on the aggregative statistics in social network,the main contributions of this research are shown as follows:Firstly,this thesis gives the definitions of the social network and the aggregative statistics and so on,and then gives the definition of the product recommendation problem based on aggregative statistics.And a product recommendation probabilistic factor graph has been used to represent the social network,which provides the data basis for the later research.Secondly,this thesis studies a product recommendation method based on aggregative statistics called PRMAS.Base on the two main steps to solve the problem,PRMAS has been divided into two section: 1)the link prediction method based on aggregative statistics(LPMAS): use the product recommendation probabilistic factor graph to mining the link between users and items of the social network which contains aggregative statistics,which means to find who like the items indeed;2)the product recommendation method based on aggregative statistics(PRML): based on the links we has predicted before,we take the interest similarity,the connection strength and the popularity into consideration to recommend the items which the users may like to the users.Thirdly,several experiments are designed and implemented.The experiment results on Amazon and CiteSeer Citation data set prove the feasibility and correctness of the newproposed method PRMAS of the paper.