Personalized Information Recommendation Considering Information Attributes And User Grades

With the rapid development of internet and information technology,we have entered an era of information overload and fragmentation.Faced with the huge amount of network information,it is difficult for users to find useful information,and is also difficult for information publishers to publish their own information in front of the user interested in it.In order to alleviate these problems,personalized information recommendation system came into being.However,the situation has features with the rapid upgrading rate of information,the change of users' interests and the contineous subdivision of information attributions and users' permissions.But traditional personalised recommendation algorithms care less or ignore about this situation,which causes the searching result is hard to meet user's demand.In order to give users better personalized information recommendation service,the above problems are the focus of this paper.The main work of this paper is:1.Research on the composition of user model.In this paper,the interception factor is introduced into Content-based recommendation algorithm,and the user model is only generated by some of the latest information,which alleviates the problem that the system can recommend the information that the user may like in the past but not currently pays attention to.2.Research on the process of similarity calculation.In this paper,the interception factor is introduced to content-based similarity collaborative filtering and behavior-based similarity collaborative filtering to get the latest browsing information of the target user.We calculate the similarity between the target user and other users,and find out the user's similar neighbor set.3.Research on the process of recommendation result generation.In this paper,we propose a novel combined recommendation algorithm——CR algorithm,which consists of two parts: one part is generated by the hybrid recommendation algorithm,and another part is generated by the user-based collabrative filtering algorithm.Thus,the required message can be obtained by filtering out news,whose published time is bigger than the current time in a certain threshold.Then it can be recommended by the hybrid recommendation algorithm.(For the situation that the messages whose published time is less than the current time in the threshold,the messages will be recommended by the user-based collabrative filtering algorithm if they meet the recommended requirements.)4.In this paper,aiming at the continuous subdivision of information attributions and users' permissions,we design a IR algorithm that takes into account both the attribute of information and the user's authority.The algorithm firstly uses the improved decision tree algorithm to classify the information on the server,and then analyzes the relationship between various subsets of information in server and the role,and then analyzes the information transmitted between the server and the role,Finally,analyzes and solves the information transfer between roles and users.5.In this paper,CR algorithm and IR algorithm are verified by experiments.The experimental results show that the CR algorithm has great advantages in F value,precision,recall rate and diversity compared with the similar algorithms.In the aspect of information classification,the improved ID3 algorithm in the IR algorithm is better than the traditional classification algorithm in the accuracy and the user's actual needs.In the aspect of information transmission,the reliability is verified by the experiment.