Websites’ Recommendation System Research Based On Folksonomy And HOSVD

Folksonomy appeared with the development of web2.0,and got a very extensiveapplication. Folksonomy pay attention to user participation,it prefer to theUser-Tag-Item ternary relationship rather than the entry-Item binary relationship ofthe traditional taxonomy. By inviting users to participate in the taggingprocess,folksonomy achieve the goal of knowledge sharing and user interaction.Folksonomy allow users to freely indicate Items they interest in. Websites constructedwith folksonomy can collect and organize labeling informations which is used todiscover users’ potential prefences which can help websites with meeting users’ needsin turn.There are many mature recommendation system is applied to the use of websitesbuild by taxonomy nowadays. One of the most classic recommended strategy is touse the cosine method to calculate neighbor similarity between users or similarityitems firstly, then use the K-nearest neighbor method to filter out low similarity itemand generate the recommendation results. With the development of folksonomy, anew ternary relation mode gradually spread on the Internet, directly using the originalrecommendation system will not be enough to better dig out the potential userpreferences. This article aimed at websites constructed by new building pattern,improve the traditional recommendation system, and put forward a (PersonalizedRecommendation System)PRS, PRS can better mining preferences. According todifferent users PRS can generate personalized Recommendation results closer to theexpectations of different users.Ternary relationship has bigger data quantity than binary relationship,accordingly, PRS recommendation systemmust deal with more data than traditionalrecommendation system, and PRS will face more serious data sparseness problem. InOrder to alleviate this problem, this article introduce HOSVD (Higher Order SingularValue Decomposition) algorithm, deal with ternary relationships of website build byfolksonomy firstly, is a good way to ease the higher-order data sparseness problem,and through this way PRS can have more excellent recommendation performance.The key contents of this article are as follows:(1) Use ternary relation data model of websites build by folksonomy to create athird order tensor space, and use the HOSVD algorithm to deal with the third order tensor, as much as possible to fill the gap data on the basis that the premise of originaldata correlation, so as to effectively reduce the data redundancy, improve theefficiency and precision of the PRS recommendation system.(2) To improve the traditional recommendation system, develop a newpersonalized recommendation system PRS, through calculating label and itemsimilarity, tag potential preferences of users and items, draw the PRV(Personalizedrecommendation value) score of each items when users retrieve something,descending order all items according to the PRV can get the recommendation resultsof PRS.(3)Using the online folksonomy data package in simulation experiment, bycomparing the PRS with several kinds of traditional recommendation system onrecommend effect and the operation time, testing performance and practicability ofthe PRS.