Research Of Recommendation System Based On Folksonomy And HOSVD

With the rapid development of Web2.0technology, the concept of folksonomy hasbeen generated. The function of tagging in folksonomy website could make users tag relatedinformation to the resources they are very interested in on this website. A tremendous plentyof tagging information that was settled and analyzed could supply the reference aboutwebsite to the new users thus to form recommend resources.At present there have been so many kinds of personalized recommendation systems inthe area of electronic commerce and the more mature of recommendation algorithm isanalyzing the similar preference degrees between item resources and neighbor users to filterinformation so as to accomplish the aim that recommending to others. But the character offolksonomy data is tags connect users and items so that calculating the similar preferencebetween neighbor users items and tags makes the system operation so much complicated, inthe meanwhile, with the increasing of website’s scale and data the issue of sparsity in datamatrix could lead the accuracy of recommend in this system decrease dramatically.Aim that issues mentioned above that part, this paper uses higher order singular valuedecomposition algorithm based on the dimensional reduction of tensor and data clustertechnology to build a new kind of personalized recommendation system according to thedata that has typical feature of folksonomy websites. The recommendation system oftagging information websites has been improved to avoid the complex process of operationand increase the recommend precision. There are several research emphases in several waysof this paper:1. The data list containing users, tags and items has been built with collecting fromclassic folksonomy websites so as to use K-means technology to cluster initial data. The aimof this step is to add relevance of data and reduce the spare parts of data group in order toprovide fundamental information to construct tensor model.2. The three-dimensional tensor space matrix could be built by ternary data clusteredwith K-means algorithm and two-dimensional SVD has an excellent feature on handling theproblem of sparsity in data matrix that could be expended in multi-dimensional tensor space.This method could delete the spare parts of data on this occasion of ensuring the completionof data construction to reduce influence of sparsity in the matrix and as the same timegenerate the result of recommendation. It could drop the data redundancy in folksonomy websites and average the accuracy up effectively.3. The comparison of this method with other classic recommendation algorithms couldbe finished through the data experiment to exam the effectiveness of this unified methodthat combine K-means and HOSVD algorithm in folksonomy websites by this paperrepresented.