Gauss-core extension dependent prediction algorithm for collaborative filtering recommendation

Abstract

Most current researches on collaborative filtering recommendation generally improve the performance of traditional algorithm within its framework. Different from these studies, the paper presents new collaborative filtering framework by introducing a new gauss core based dependent function and an extension classification method. It builds matter-element classification by statistical information from input data, and recommends results by calculating dependent degree between characteristics of classification set and that of target object. Since it uses the dependent degree between user and classification set rather than the similarity between users or items in traditional algorithms, this method can be applied for various kinds of attributes value and can naturally shield data sparsity problem. Furthermore, it not only copes with several limitations of the traditional recommendation algorithm, but also resolves intrinsic problems of traditional extension dependent algorithm. The data experiments show the robustness, higher efficiency and better performance of our new algorithm.