Abstract: An ad system is described for providing ad selections in response to an ad presentation opportunity. The ad system can use multiple content providers to generate multiple sets of ad candidates. The content providers can apply different ad selection paradigms in generating their sets of ad candidates. The paradigms may act on different aspects of a context pertaining to the ad presentation opportunity. A chooser module and filtering module can cull the set of ad selections from among the plural sets of ad candidates. At least one content provider uses machine learning functionality in generating ad candidates, such as a Naïve Bayesian approach, an affiliate similarity approach, etc. Various content providers also find application in a stand-alone mode.