The articles mentioned ad relevance score, but when compute the ad rank score, we are using ad predicted score. Is ad relevance score same as ad predicted score? If not, where do we use ad relevance score?
Ad relevance and prediction scores are different.
Ad relevance score is a part of the ad selection process. The ads selection component selects all the ads that match the targeting criteria (user demographics and query) and uses a simple model to predict the ad’s relevance score.
These selected ads are then passed to the ads prediction component. The ads prediction component goes over the selected ads (which were selected using the ad relevance score) and uses a highly optimized ML model to predict user engagement through a precisely calibrated score called the ad prediction score.
So in summary, ad relevance score is used to select relevant ads from a pool of ads. Ad prediction score predicts user engagement for an ad.