Using estimated ad qualities for ad filtering, ranking and promotion

1. A method implemented by one or more processors of a computer system, comprising:

obtaining, using one or more processors of the computer system, ratings associated with a first group of advertisements, where the ratings, which include a quality of the first group of advertisements, are received from human raters;

observing, using one or more processors of the computer system, multiple different user actions associated with user selection of advertisements of the first group of advertisements;

deriving, using one or more processors of the computer system, a probability model using the observed user actions and the obtained ratings, where the probability model includes a probability function that specifies a probability that an advertisement, of a second group of advertisements, is of a certain quality as a function of multiple different types of user actions, where deriving the probability model comprises:

using at least one of logistic regression, regression trees or boosted stumps to generate the probability model;

using, by one or more processors of the computer system, the probability model to estimate quality scores associated with the second group of advertisements;

calculating, using one or more processors of the computer system, a combination of the estimated quality scores and click through rates associated with advertisements in the second group of advertisements;

filtering, using one or more processors of the computer system, the second group of advertisements based on a comparison of the combination of the estimated quality scores and click through rates to a threshold to generate a subset of advertisements from the second group of advertisements; and

providing, to a user, the subset of advertisements from the second group of advertisements.

2. The method of claim 1 , where calculating the combination of the estimated quality scores and click through rates further comprises:

calculating a combination of the estimated quality scores, the click through rates, and cost-per-click (CPC) associated with advertisements in the second group of advertisements; and

further comprising:

ranking the advertisements of the subset of advertisements based on the calculated combination of the estimated quality scores, the click through rates and the CPC to order the subset of advertisements.

3. The method of claim 2 , further comprising:

promoting one or more of the subset of advertisements based on the calculated combination of the estimated quality scores and click through rates.

4. The method of claim 3 , further comprising:

positioning the promoted one or more advertisements of the subset of advertisements in a prominent position on a document; and

positioning unpromoted advertisements of the second group of advertisements in a less prominent position on the document than the promoted one or more advertisements.

5. The method of claim 4 , where providing the subset of advertisements to a user comprises:

providing the document to the user.

6. The method of claim 2 , further comprising:

providing the subset of advertisements in the ranked order to the user.

7. A system, comprising:

means for obtaining ratings associated with a first group of advertisements, where the ratings, which include a quality of the first group of advertisements, are received from human raters;

means for observing multiple different user actions associated with user selection of advertisements of the first group of advertisements;

means for deriving a probability model using the observed user actions and the obtained ratings, where the means for deriving a probability model further comprise:

means for using at least one of logistic regression, regression trees or boosted stumps to generate the probability model;

means for using the probability model to estimate quality scores associated with a second group of advertisements;

means for calculating a combination of the estimated quality scores and click through rates associated with advertisements in the second group of advertisements;

means for filtering the second group of advertisements based on a comparison of the combination of the estimated quality scores and click through rates to a threshold to generate a subset of advertisements from the second group of advertisements; and

means for providing, to a user, the subset of advertisements from the second group of advertisements.

8. The system of claim 7 , where the means for calculating the combination of the estimated quality scores and click through rates further comprises:

means for calculating a combination of the estimated quality scores, the click through rates, and cost-per-click (CPC) associated with advertisements in the second group of advertisements; and

further comprising:

means for ranking the advertisements of the subset of advertisements based on the calculated combination of the estimated quality scores, the click through rates and the CPC to order the subset of advertisements.

9. The system of claim 8 , further comprising:

means for providing the subset of advertisements in the ranked order to the user.

10. The system of claim 8 , further comprising:

means for promoting one or more of the subset of advertisements based on the calculated combination of the estimated quality scores and click through rates.

11. The system of claim 10 , further comprising:

means for positioning the promoted one or more advertisements of the subset of advertisements in a prominent position on a document; and

means for positioning unpromoted advertisements of the second group of advertisements in a less prominent position on the document than the promoted one or more advertisements.

12. The system of claim 11 , where the means for providing the subset of advertisements to a user comprises:

means for providing the document to the user.

13. A computer-readable memory device that stores computer-executable instructions, comprising:

one or more instructions for obtaining ratings associated with a first group of advertisements, where the ratings, which include a quality of the first group of advertisements, are received from human raters;

one or more instructions for observing multiple different user actions associated with user selection of advertisements of the first group of advertisements;

one or more instructions for deriving a probability model using the observed user actions and the obtained ratings, where the probability model includes a probability function that specifies a probability that an advertisement, of a second group of advertisements, is of a certain quality as a function of multiple different types of user actions, where the instructions for deriving the probability model comprising:

one or more instructions for using at least one of logistic regression, regression trees or boosted stumps to generate the probability model;

one or more instructions for using the probability model to estimate quality scores associated with the second group of advertisements;

one or more instructions for calculating a combination of the estimated quality scores and click through rates associated with advertisements in the second group of advertisements;

one or more instructions for filtering the second group of advertisements based on a comparison of the combination of the estimated quality scores and click through rates to a threshold to generate a subset of advertisements from the second group of advertisements; and

one or more instructions for providing, to a user, the subset of advertisements from the second group of advertisements.