Collaborative Filtering Approach Based On User Preference And Social Attribute Similarity

The need for recommendation algorithms is very early,before machine learning was fully popular.Especially in the period of rapid development of Internet,the network is filled with a large number of useless information,resulting in serious information overload,which makes the demand for recommendation system more and more obvious.The recommendation system connects users' needs with information.Even if users can find valuable information quickly,the information can be pushed to customers who need it reasonably,thus reducing the cost of both parties.Common recommendation algorithms often need sufficient information to have a good recommendation effect.Therefore,when faced with sparse data and unable to reach the requirement of accurate calculation,the final recommendation accuracy will be affected,thus bringing bad experience to users.A collaborative filtering algorithm based on user preference and social attribute similarity is proposed to solve the cold start problem caused by sparse data.First,user rating matrix and project attribute matrix are used to get the attenuation function based on improved time user preference matrix,this method can reduce the data sparse sex brings the problem of difficult to calculate similarity between users,and according to user preference matrix can better classifying users,using the user preference matrix using clustering method to classify the user,then according to the clustering algorithm for user label,and the social attribute of stacking method is used to training integration model,which can in new users into the system,a more accurate classification,and then get new users of the nearest neighbor set,Finally,the top-n algorithm is adopted to generate the final recommendation list and generate recommendations.This method can make up for the problem that new users or new projects have no feedback information in the system,and effectively alleviate the problem of cold start and data sparsity.Finally,through several experiments,the improved algorithm was compared with thetraditional cold start solution by using such indexes as prediction accuracy,recall rate and accuracy rate,etc.,and the results proved that it could effectively improve the recommendation quality.