Research On Personalized Recommendation Algorithm Based On User Attributes And User Preferences

Currently,with the rapid development of network technology,people can find many different kinds of goods on the Internet.However,because of the excessive information,users can not quickly find out valuable and interesting information from massive data.In order to solve the "information overload" situation,the recommenddation system can recommend various types of items according to the characteristics and preferences of different users.It is widely used in music,video,e-commerce,location services and other scenarios.Particularly,collaborative filtering is one of the most popular recommendation algorithms.The problem of cold start and sparsity exist in the traditional collaborative filtering algorithm.What's more,users' interest may change over time due to long time interval to score the project,causing this recommendation algorithm didn't work well.Based on the traditional collaborative filtering algorithm,this paper made further research.And the main work of this paper was as follows:(1)Considering the cold start of users in traditional collaborative filtering algorithm,this paper used the recommendation algorithm based on user clustering to study which clusters user attributes by K-means algorithm.The traditional recommendation algorithm performed the score prediction according to the users' scores of the product,but the newly registered users did not have any score record.Therefore,this paper performed the recommendation list by predicting the score according to the users' attributes,and then the target users were recommended.Clustering was on the basis of users' registration information,at a certain level to alleviate the users' cold start problem.What's more,even if there was no score value,it could give recommendations to the target users.(2)Aiming at the problem that users' interests would change from time to time,this paper introduced time weighted.When calculating users' preferences,considering the sparsity of data,the user prediction score was calculated by pearson correlation coefficient in clustered users.And then the user-item score matrix was recalculated by weighting users' preferences according to time function.(3)Combining the user-clustering-based recommendation algorithm and user preference time weighted recommendation algorithm,the simulation ran with MovieLens 100 K data set.The experimental results showed that the MAE value of the proposed algorithm was lower than that only considering user time weighted.