Research On Personalized Recommendation Method Of Multimedia Resources

In order to help users quickly find the resources they are interested in,this thesis designed a personalized recommendation algorithm for multimedia resources.The thesis used the related dataset from an internship program named Duo Yue reading platform.The research is carried out in the following three aspects: user rating strategy,user characteristic model and cold-start problem in collaborative filtering recommendation.The main contents of this thesis are as follows:(1)User Rating ProcessIn order to generalize the user rating,the main idea of this thesis is to use the user's behavior feedback as the resource evaluation index from users,and assign the corresponding weights to users' different behavior categories.The weight is calculated by the entropy method.The process includes: defining the regular tag firstly,and then determining the feature model of the resource and the representation of user's operation,and finally obtaining the final user-tag score based on the resource-tag association matrix,the user-resource association matrix and the resulting behavior weight.(2)User Feature ExtractionBy the existing recommendation algorithms,the user feature is usually not accurately discovered in relevant case studies.The main idea of this thesis is to analyze user characteristics by using user's regular tag characteristics,behavior characteristics and time-related characteristics,and finally establish the user's characteristics model.(3)Collaborative Filtering Algorithm ImprovementBased on the improved user feature extraction method,the collaborative filtering algorithm is used to find similar users and calculate the recommendation results.Considering the problem of the scale of target user's neighbor set,the result of the final recommendation in this thesis is divided into two parts: recommended multimedia resources depending on the target user characteristic and the similar user,and the relationship between them can be adjusted by setting parameters.The effectiveness and accuracy of the proposed method are proved by experiments.(4)Research on Cold Start Problem of Collaborative Filtering AlgorithmIn order to solve the problem of “cold start” for new users,the main idea of this thesis is to analyze the four basic user attributes: age,sex,profession and region.A decision tree classifier is created by the basic information of existing users.When a new user enters the system,we use the decision tree to classify him into certain category,and the new user's feature model is built according to similar existing user characteristics model,so as to complete the recommendation.The effectiveness and accuracy of the algorithm are proved by experiments.Finally,the feasibility of the multimedia resources recommendation method proposed in this thesis is verified.It mainly includes requirement analysis,design and implementation of recommendation functions.