Linguistic selectors and their optimization

Mining through vast arrays of heterogeneous data (no matter where they come from) is a challenging and rewarding pursuit. To make the findings meaningful, the methods of data mining need to be presented to the end-user in a highly intelligible manner. The role of information granules is to cast data mining in the setting of some meaningful nonnumeric entities exhibiting a well-defined semantics. We propose a concept of a linguistic selector and view it as a focal element of data mining. The linguistic selectors are weighted and-combinations of the individual linguistic terms: fuzzy sets or fuzzy relations defined over features (variables) existing in the problem. When operating on a database, they are matched with the individual records and return a certain level of compatibility (matching). The fundamental aspect of such linguistic selectors deals with their relevance in terms of their semantics and statistical relevance. We quantify these two features and propose an optimization problem leading to the design of the meaningful linguistic selectors.