Researches On Top-N Collaborative Filtering Recommendation Algorithms Based On SVM

The rapid development of computer and Internet,especially the wide application of mobile Internet technology,the way for people accessing and sharing information was complete changed by Internet.The various needs of users were satisfied by vast amount of web information,and the information can give users great support and help.However,the over-expansion problem about web information has happened due to its diversity and dynamics,the over-expansion problem who is named information overload,which makes it becoming increasingly hard for people to find relevant information quickly and accurately from the vast web information.Recommender systems have been introduced in recent years to help people in retrieving potentially useful information or products.Meanwhile,recommender system is a hot research topic in the field of data analysis and mining in Big Data,it has great value in theory and application,which gains much attention from researchers at home and abroad.The main goal of recommender systems is to find some recommendations that match users' interests,and find some suitable ways to display them for users.The core target of Top-N recommendation is to find a set of N items to recommend for users,the small amount of accurate recommendations more conform to users' choice habits in comparison to other methods,so Top-N recommendation has become a hot research problem.However,most classical Top-N recommendation algorithms don't make full use of negative feedback information and user distribution information.We proposed several effective solutions to the relevant problems,the main contents of this thesis include the two following aspects.(1)Researches on Top-N recommendation algorithms based on user positive and negative feedback information.Most classical Top-N algorithms need to rank all the items in order to identify a set of N items that will be of interest to a certain user.To make full use of positive and negative feedback information,this thesis proposes SVM Collaborative Filtering approach based on Positive and Negative Feedback(PNF-SVMCF)recommendation algorithm.This method can reduce the number of items which need to be sorted by using negative feedback information,and promote the recommended efficiency.At the same time,it can eliminate interference of those items which user dislike,and improve recommendation performance.(2)Researches on PNF-SVMCF based on users' granular distribution information.Considering a personal relationship between users is different,we determine to optimize recommendation model by using users' granular distribution strategy.To verify this strategy is useful,we use this strategy to optimize SVM classification efficiency,and presents SVM training process based on granular distribution strategy(GDSVM).It selects representative points according to the distribution of positive and negative samples in the granules.Then it trains SVM classification model by these samples.Next,we presents PNF-SVMCF based on user granular distribution strategy(PNF-GDSVM).We cluster all users at first,then reconstitute rating matrix through the different granular distribution information whether the target user is exist in a user granular,and accomplish the Top-N recommendation at last.To solve the problem that most classical Top-N recommendation algorithms don't make full use of negative feedback information and user distribution information,we proposed the PNF-SVMCF and PNF-GDSVMCF model,which can improve recommendation precision and speed.At the same time we using the granular distribution strategy to SVM classification model,and we proposes GDSVM model which has high learning efficiency.The obtained results of the thesis will be significant for application researches of Top-N recommendation algorithm and SVM classification model.