Research On A Publishing Resource Recommendation Algorithm Based On Optimized Tag Features

For the publishing industry with a lot of resources,personalized recommendation is an effective way to solve the problem of information overload.In various recommendation algorithms,the matrix factorization algorithm has achieved good recommendations,but there are also some problems.Based on the matrix factorization algorithm,this thesis uses the tag feature to analyze the influence relationship between users and resources and integrates the relationship into the matrix factorization algorithm.And we studied the problems in the traditional recommendation algorithm,such as matrix sparsity problem,inaccuracy in user resource feature mining,cold start problem and recommendation efficiency decrease as resources increase.The main contents of this thesis are as follows:(1)Improvement of user and resource feature extraction method.Feature extraction of users and resources is based primarily on tag characteristics.By collecting user behavior information,the relationship between the user and the resource can be found.According to the standardized tag of the resource itself,the relationship between the resource and the tag can be obtained.But both of the relationship matrices have obvious sparse problems.In order to find the importance of tags for users and resources,this thesis uses the iterative method to weight the tag feature matrix to extract more accurate feature matrices of user and resource.Finally,the PCA dimension reduction technique is used to reduce the dimension of the relation matrix to alleviate the data sparse problem.(2)Improvement of probabilistic matrix factorization algorithm.Aiming at the problem of ignoring the influence relationship between users and resources in the traditional probabilistic matrix factorization algorithm,this thesis uses the user and resource feature matrix to find the neighbors of users and resources,and then integrates the similar neighbor relationships of users and resources into the probabilistic matrix factorization algorithm.Experiments show that the improved algorithm improves the recommendation accuracy.(3)Cold start problem research.In this thesis,the decision tree model is constructed by using the user information such as gender,age,access time and location to solve the cold start problem in the recommendation algorithm.When a new user enters the system,the system matches the user information with the decision tree model to obtain a list of recommended resources.(4)Interactive recommendation framework research.The recommended efficiency will gradually decrease and user interest may change in a short time as the recommended resources increase.In order to solve the problem,this thesis introduces an interactive recommendation framework,which obtains the user's current interests through the interaction between the user and the tag,and narrows the resource set according to the tag,thereby obtaining the recommended resource.Finally,this thesis conducts specific experiments on the research content and existing problems,and uses the evaluation indicators to analyze the experimental results to verify the improvement effect of the algorithm.