Abstract: The present disclosure, in some embodiments, describes a system for classifying members of a collection of texts into clusters to generate merged data collections. A member text can range from a single document to the contents of a column in a database table. The classification may indicate and/or provide an estimation as to which documents or columns are most closely similar to each other, without making any assertion about the actual contents of the document or column. In some embodiments, a system may include counting some characteristic of the text. The characteristic may be chosen such that each text produces a set of counts. A statistical measure is then applied to determine the similarity of sets of counts associated with each pair of texts.