Collaborative Filtering Recommendation Algorithm Based On User Multi-dimensional Similarity

With the popularity of handheld devices and desktop computers,the network has become an indispensable part of people's life.Many social activities conduct through the Internet,and the network information is ever-accelerating pace,using information rationally on the Internet,the first thing to do is to filtering and sorting.Recommender system built a bridge between users and in network information,it facilitates the people's lives.From 90 th early last century to now,recommender system increasingly matures,this paper according to user information starting,analysis user property and network behavior from many angles,research how using and Mining user information then improved collaborative filter recommender system.The main work is as follows:Firstly,this paper introduced recommender system in both at home and abroad of development status,induce and order collaborative filter recommender algorithm,and on its for principle analysis,on which related of key technology for analysis.Secondly,The similarity calculation.As recommender system of core module,similarity of calculation directly affect recommend results,on recommend precision,and coverage,and recall rate has directly of relationship,The key to optimize recommender system is optimize similarity calculation,This paper through research similarity calculation principle on traditional similarity calculation method for improved,using individual differences between users for calculating project of preference difference,then calculation out users similarity.By making experiments in open data set and continuously adjust the parameters,making that the resulting precision has the ideal level.Thirdly,reviews text mining and utilization.Relatively conventional scoring or rating,reviews more representative of the user's visual experience of natural language text reviews will undoubtedly contain a large amount of user information and user preferences.Using crawlers,we get a large number of hotels the reviews text,and Chinese reviews text processing system,contains user information extraction and quantification of emotional reviews text,users will get emotional tendency analysis and calculated on the similarity between users of the data.Fourthly,Multi-dimensional Similarity algorithm.Calculation of similarity should not be based on a user attribute or behavior only,all user information which can influence the choice should be fully taken into account in order to calculate more accurate user neighbors for making precise recommendations.This paper collects user's attributes,user rating set,sentiment and other elements contained in the user reviews,assign it different weights involved in the calculation of similarity.Recommendations are constantly adjusted the weights to select the optimum results.From the different dimensions to calculate the user similarity,focusing on the information included in the emotional tendencies contain in user reviews,resulting in more accurate user neighbor.Experiments and comparisons show that this paper can provide better results for users.