Abstract: A parameter-selection technique that allows preferred permutations to be rapidly identified based on real-world user behavior is described. In particular, a set of web pages that include different permutations of the set of parameters are generated. Then, at least a subset of these web pages is provided to the users in response to their requests. For example, the requests may be associated with user selections of search results, which are associated with search queries provided by the users to a search engine. While the users interact with the subset of the set of web pages, their actions and the associated context (with respect to the different permutations) are tracked. Next, the tracked user actions are used to identify the web pages associated with the preferred permutations of the set of parameters.