Data mining, particularly pattern mining and cluster mining, aims to find interesting and characteristic local structures in data with enumerational approaches, not to miss important ones. However, a big difficulty is on the huge number of solutions so that we can not identify which solutions are really important. This difficulty is a great barrier on the use of pattern mining and cluster mining in practice. In this talk, we propose a new approach called "data polishing" that takes an approach totally different from existing approaches. The idea is to modify the data by feasible hypothesis, so that ambiguities will be distinguished. The result is fine, and even for graph visualization we can see many clusters clearly, in SNS networks.
