Constructing the compressed description of dataset by the function of rival similarity

We argue that the general aim of data mining consists in constructing some simplified compressed description of information. The Function of rival similarity (FRiS-function) is proposed as a new ternary similarity measure between objects instead of a binary one. Quantitative estimation of the compactness of datasets, basing on FRiS-function, allows constructing new more effective compressing algorithms of data mining. Some examples are described of the algorithms testing on real and model tasks.