We present a framework that enables the use of traditional feature selection algorithms in a new context - for building a set of subsets of specified properties. During the course of search individual items are added/removed to/from one of the subsets in the subset system one at a time to maximize an overall criterion. Different tasks of prototype search type can be solved in this alternative way depending on suitable criterion definition. The usability of the concept is shown on keyword extraction example. Further possible applications are suggested.