Recommendation systems, best known for their use in e-commerce or social network applications, predict users' preferences and output item suggestions. Modern recommenders are often faced with many challenges, such as covering high volume of volatile information, dealing with data sparsity, and producing high-quality results. Therefore, while there are already several strategies of this category, some of them can still be refined. Association rules mining is one of the widely applied techniques of recommender implementation. In this paper, we propose a tuned method, trying to overcome some defects of existing association rules based recommendation systems by exploring rank correlations. It builds a model for preference prediction with the help of rank correlated associations on numerical values, where traditional algorithms of such kind would choose to do discretization. An empirical study is then conducted to see the efficiency of our method.
展开▼
