A Collaboration Filtering Recommendation Algorithm Based On Fusing User Rating And Item Attribute

With the Internet gradually permeating in human life,the lifestyles of people has gradually changed to intelligent.The intelligent product meets the individual need of user.And the recommendation system provides the user with product information according to the individual need of the user.The recommendation system provides users with the needed information actively,helps users make better decisions and solves the problem of the user for the "buy what".Now recommendation system has been successfully applied to social network,e-commerce,search engine,advertising recommendation and other fields.The quality of the recommendation system mainly depends on the recommendation technology.As a recommendation technology,collaborative filtering recommendation that is applied and researched hottest has its unique advantages,but it also exists some problems such as data sparse,cold start and scalability and other problems.In this paper,on the basis of the recommendation technology,in order to solve the problem of low accuracy and poor scalability of the recommendation system,studying the collaborative filtering recommendation algorithm thoroughly,and the collaborative filtering recommendation algorithm based on fusing user rating and item attribute is further proposed.Aiming at the problem of the poor scalability and sparse scoring data,the user characteristic model and item attribute model are designed,and an algorithm to update the user's rating matrix is proposed in this paper.In order to reduce the search space of the neighbor set of the target user or the target item and reduce the computational complexity of the system,this paper respectively clusters the user characteristic and item attribute with the k-means clustering algorithm.The algorithm predicts the value of user-item rating that is not rated by using the item-based collaboration filtering algorithm on the basis of clustering the item attribute,and then fills the default value of the user-item rating matrix to update the user's rating matrix.The algorithm effectively mitigates the influence of sparse scoring data.Aiming at the problem of low accuracy of the recommendation system,on the basis of updating the user rating matrix,a similarity calculation method based on fusing user rating and item attribute is proposed in this paper.The method takes into account the entropy of the user rating difference and the similarity of the item rated attribute.Then,based on the clustering user characteristic,the similarity calculation method is used to find the neighbor set of the target user,which improves the accuracy of the neighbor user searching.Finally,the experiment is done in the following two conditions:the first,setting the default value of the user-item rating to 0;the second,filling the default value of the user-item rating with using the updating user rating matrix algorithm.The validity of the similarity calculation method in this paper is verified by comparing with the other similarity calculation methods,and the recommendation quality of which is better under the second condition.