A system for generating normalization candidates for a search query includes a database for storing session logs with each session log including query data and a processor in communication with the database and configured to execute computer-readable instructions causing the processor to analyze session log data to generate sets of query reformulations for a plurality of search queries, select one of the sets containing a normalization candidate that matches the search query, filter the selected set of reformulations, tie the candidates in the selected set to a category, compare the categories of the candidates, remove at least one reformulation from the selected set when the category of one candidate is uncommon with the category of the other candidate, and store the remaining candidates in the database. A method and one or more non-transitory computer-readable storage media for generating stemming pairs for a search query are also disclosed.