Research On Interactive Recommendation Method For Publishing Resource Based On Tag Association Analysis

With the rapid development of Internet technology,how to satisfy users' personalized knowledge service needs is an urgent problem to be solved in publishing industry which has abundant resource.In this thesis,a personalized recommendation algorithm for publishing resources is particularly designed based on the resource characteristics of the publishing industry,at the same time we have researched the problems in the recommendation algorithm,such as score matrix sparsity,recommendation performance decrease with the increase of resources,cold start and so on.The main contents of this thesis are as follows:(1)Improvement of user feature extraction method.In this thesis,we improve the users' feature extraction method in traditional recommendation algorithm,with full consideration of users' tag characteristics,behavior characteristics and time characteristics,and build the users' interest characteristic vector.Based on the standardized tag,we get users' tag characteristics through users' action feedback and interaction operation on the resources,and define the tag weights by different users' behavior feedback,meanwhile through the time forgetting curve consider users' tag feature offset.(2)Improvement of probabilistic matrix factorization algorithm.In this thesis,the problem of neglecting the relationship between users and resources in the traditional probabilistic matrix factorization algorithm is improved.First,we get the user and resource feature by using standardized tag and find the neighbor of users and resources,then we integrate the relationship of the similar neighbor sets into the probabilistic matrix factorization to reduce the sparsity of the scoring matrix and improve the accuracy of the algorithm.The validity and accuracy of the improved probabilistic matrix factorization are proved by a large number of experiments.(3)Interactive recommendation framework.In this thesis,an interactive method based on tag association analysis is proposed,and we have designed an interactive recommendation framework,which can reduce the resource selection set by user interaction,locate the recommendation results,and improve the efficiency of the recommendation algorithm.In the process of user interaction,we analyze the tag through the resource-tag matrix,provide better tag attributes for users to choose and optimize the dividion of resource alternatives set.Interactive experiments demonstrate that the interactive approach improves the performance of the recommendation.(4)Research on cold start problem.In order to solve the cold start problem of new users in recommendation process,in this thesis we analyze four basic information attributes of users: age,gender,reading time and reading place.A decision tree classifier is created for the basic information of users.When new users enter the system,matching the decision tree classification model,initializing the recommended topic resources.Meanwhile,in the interactive framework,we can quickly get users' needs through tag interaction.The experiment based on the actual dataset proves the validity and accuracy of the algorithm.Finally,an interactive recommendation method based on tag association analysis is verified in this thesis,at the same time,we have carried out the requirement analysis,the module design,the frame design and the system implementation.