Research On A New Collaborative Filtering Recommendation Algorithm

With the development of the Internet,the recommendation algorithm has been applied to many fields,collaborative filtering recommendation algorithm is a classical and widely-used recommendation algorithm.However,the traditional collaborative filtering recommendation algorithm is facing many problems,the most serious of which is the cold-start problem,the problem of data sparseness and scalability problem.Aiming at these problems,this paper makes some improvement on the traditional collaborative filtering recommendation algorithm.First of all,aiming at the problem of data sparseness,we propose a collaborative filtering recommendation algorithm based on item similarity learning.Firstly,we calculate the similarity matrix of all items according to the measurement of item attribute similarity,and then select the K most similar items of the target items as the initial neighbor set;then set the score vector of the target item in the training set as the desired output.Input the score vectors of K adjacent items of the target item to the RBF neural network for learning and training to get the item similarity model;then input the score vectors of K adjacent items of the target item in the test dataset into the training model,finally output the predictied score vector of target item.Aiming at the cold-start problem of new item,we calculate the attribute similarity of the new item and other items,and then select the K most similar items as the initial neighbor set,and calculate the predictied score vector of the new item.Finally,we select the users who have scored more than or equal to 3 of the target item and the scores are in the top N list,then recommend the target item to these users.Second,aiming at the problem of data sparseness and scalability,this paper proposes a collaborative filtering recommendation algorithm based on social network and label.This algorithm will combine direct trust,familiarity and the interest preference similarity reflected by the label information between the target user and his friends to calculate the K most similar friends as the neighbor set,so as to recommend love items for target users;then,aiming at the cold-start problem of new user,we put forward a model based on Naive Bayes algorithm.It makes use of the Naive Bayes algorithm to classify the users in training dataset,and divides the new users into their own categories,that is,to find out the new user's favorite item type,and then recommend the highest rated N items in this type of item to the user.Finally,we have achieved the collaborative filtering recommendation algorithm based on item similarity learning on the Movielens dataset,the cross experiments show that this algorithm shows good performance in dealing with sparse data,and get more accurate recommended result;implement the collaborative filtering algorithm based on social network and label information on the Last.fm dataset,and this algorithm has good accuracy and high efficiency compared with the traditional algorithm and some classical algorithms.Finally,the cold-start problems of new user and new item are verified on the Movielens dataset,the experiments show that the algorithms solve the cold-start problems to a certain extent.