Research On Recommender Systems Based On Label

Internet Plus has generated a huge wave of information,it provides a wider choice for the user,at the same time increases the difficulty of information filtering,and reduces the rate of information utilization.The purpose of the recommendation system is to help users quickly find the required information,and find a high quality and high value of resources combined with the characteristics and interests of the user.Enhancing user experience by reducing the negative impact of user to contact duplication or irrelevant information.Previous recommendations used to set score or vote as a measure of the user's attitude towards resources,but this kind of data often has a too strong dependence for the resource ontology,these indicators will lose value when the resources disappear.The introduction of the mass tagging method,bring another way of resource evaluation for the recommendation system.Tags has the features of simply use and spread easily,It can not only reflect the differences of resources,but also reflect the personal preference and characteristics of the behavior.This paper studies the recommendation system based on the label,and analyzes the application of three different kinds of recommendation algorithms in the tag recommendation system.On the basis of the traditional algorithm,introducing the association rules mining.And reduce the data sparsity of the matrix by using the K-means clustering method.Based on the implicit rule of mass user behavior,to improve the accuracy of the recommendation system and the coverage of the results of the project type.The main research work is as follows:Firstly,a recommendation method based on association rule mining is proposed.Analyze the user's behavior,extract and process the user's historical behavior.After the label cleaning,clustering,the formation of a personalized user tag library will be generated,as well as a tag cloud representing each user's tag library.Then,the association rule mining is introduced.All tags that appear in the user's tag cloud,a valid transaction will be considered between each other.Finding frequent item sets from those tags,finally,those with strong association rules are retained by setting the confidence level for rule filtering.At last,the label recommendation result will accord the rules selected,the resources represented by these tags will be recommended to the user at the same.Secondly,the personalized recommendation method for the library system is studied.There are a wide variety and huge resources of book.The use of books has the characteristics of continuity and hierarchy,resources recommended by the library system should be more systematic.Through carrying multi dimension division on the user profiles in the library system.Record reading behavior changes in the time interval,detect user interest drifting degree.Combined with labeling rules,more individual rules which with low support degree but high confidence degree and general rules which with high degree both in support and confidence will be found.Finally,according to the requirement of target users and the matching degree of the former parts,the Top-N is chosen as the recommendation.In the end,this paper Designs experiment with using the data of Douban.com.After comparing with the traditional evaluation index system of the recommendation system,The model is verified with certain usability and validity.