Yahoo Labs has active scientific efforts in modeling both ad clicks and ad conversions for the contextual advertising system.

In contextual advertising, an ad click refers to the event in which a user clicks on an advertisement that has been shown on a content page. It is often useful to know the probability with which the click will happen, or Pr(click|page, ad, user). We estimate this probability by applying machine learning and statistical techniques to an enormous amount of historical click data. Using certain statistics of this data, we can learn that certain ads are very likely to be clicked on certain pages. Furthermore, we can also compute the extent to which certain kinds of term matches between the page and ad will contribute to the click probability. The contextual advertising production system currently uses the click probability as a factor when ranking the advertisements on a content page.

An ad conversion refers to the event in which the user clicks on an advertisement, views the ad's landing page, and eventually performs a transaction on the web site of the advertiser, such as an item purchase. It is an important goal for advertisers to get more conversions from their online ads, and to get them at a low average cost. We are currently investigating a number of strategies in which we apply machine learning techniques to historical conversion data in order to help our advertisers achieve this goal.