Strategies for determining the value of advertisements using randomized performance estimates

• US 8,571,930 B1
• Filed: 10/31/2005
• Issued: 10/29/2013
• Est. Priority Date: 10/31/2005
• Status: Active Grant

• Current Assignee: A9.com Incorporated (Amazon.com, Inc.)
• Original Assignee: A9.com Incorporated (Amazon.com, Inc.)
• Inventors: Galperin, Viatcheslav
• Primary Examiner(s): Van Bramer, John
• Assistant Examiner(s): GATLING, STACIE D
• Application Number:
  US11/263,648
• Time in Patent Office: 2,920 Days
• Field of Search: 705/14, 705/14.43, 705/14.52
• US Class Current: 705/14.43
• CPC Class Codes: G06Q 30/0242 Determining effectiveness of advertisements
• CPC Class Codes: G06Q 30/0273 Determination of fees for advertising

Abstract

Strategies are described for conducting an advertising campaign using a cost-per-transaction (CPT) pricing model. In this model, the advertiser is charged when an end-user takes some express action in response to viewing the advertisement, such as by clicking on the advertisement, purchasing the advertised asset, performing a registration procedure, and so forth. Various solutions allow for the computation of the expected value of a CPT advertisement when there is a scarcity of data pertaining to the actual performance of the CPT advertisement.

Claims

1. A computer-implemented method of forming a performance estimate for an advertisement, from which to compute the advertisement'"'"'s expected value, the method comprising:

under control of one or more computer systems configured with executable instructions,

receiving performance data associated with the advertisement;

forming a probability distribution of possible performance estimates based, at least in part, on the received performance data associated with the advertisement; and

randomly selecting the performance estimate from the probability distribution of possible performance estimates, such that the performance estimate includes a randomized component.

2. The method of claim 1, wherein the advertisement is priced using a cost per transaction (CPT) model, in which a cost-accruing event occurs when the advertisement induces a transaction.

3. The method of claim 2, wherein the CPT model is implemented as a cost per click (CPC) model, wherein the transaction is defined as a click event.

4. The method of claim 1, wherein the probability distribution is a normal distribution.

5. The method of claim 1, wherein a plurality of techniques are used to compute the probability distribution, wherein the method further comprises collecting performance data pertaining to the suitability of each of said techniques, and selecting a technique deemed most suitable.

6. The method of claim 1, further comprising:

determining whether there is a deficiency of performance data for the advertisement; and

if there is a deficiency, forming the probability distribution so that it is artificially elevated, to allow the advertisement to effectively compete with more mature advertisements.

7. The method of claim 6, wherein the probability distribution is characterized by at least one parameter, wherein the probability distribution is artificially elevated by elevating said at least one parameter.

8. A computer-implemented method of ranking advertisements, comprising:

under control of one or more computer systems configured with executable instructions,

identifying an advertisement-triggering event;

identifying a set of K candidate advertisements that have a bearing on at least one characteristic associated with the advertisement-triggering event;

estimating the value of each of the set of K candidate advertisements; and

selecting a subset of N advertisements from the set of K candidate advertisements based on the estimated value of the K candidate advertisements,

wherein the estimating of the value of the K candidate advertisements comprises, for each advertisement of the set of K candidate advertisements;

forming a probability distribution of possible performance estimates based on performance data of the advertisement; and

randomly selecting a performance estimate from the probability distribution of possible performance estimates to use as the estimated value of the advertisement.

9. The method of claim 8, wherein one or more of the set of K candidate advertisements are priced using a cost per transaction (CPT) model, in which a cost-accruing event occurs when an advertisement induces a transaction.

10. The method of claim 9, wherein the CPT model is implemented as a cost per click (CPC) model, wherein the transaction is defined as a click event.

11. The method of claim 8, wherein the selecting of a subset of N advertisements from the set of K candidate advertisements comprises:

ranking the set of K candidate advertisements by their respective estimated expected values; and

selecting the subset of N advertisements as the top N ranking advertisements.

12. The method of claim 8, wherein at least two of the set of K candidate advertisements are priced using different respective pricing models, and wherein the ranking is configured such as to accommodate normalized comparison between said at least two advertisements.

13. The method of claim 8, further comprising pricing the subset of N advertisements.

14. The method of claim 8, further comprising:

determining whether there is a deficiency of performance data for said at least one advertisement; and

if there is a deficiency, forming the probability distribution so that it is artificially elevated, to allow said at least one advertisement to effectively compete with more mature advertisements.

15. A system comprising:

a memory and a processor;

advertising management functionality stored in the memory and executable on the processor for evaluating the value of advertisements, comprising;

logic configured to identify a set of K candidate advertisements that have a bearing on at least one characteristic associated with an advertisement-triggering event;

logic configured to estimate the value of each of the set of K candidate advertisements; and

logic configured to select a subset of N advertisements from the set of K candidate advertisements based on the estimated value of the K candidate advertisements,

wherein the estimating of the value of the K candidate advertisements comprises, for each advertisement of the set of K candidate advertisements;

forming a probability distribution of possible performance estimates based on performance data of the advertisement; and

randomly selecting a performance estimate from the probability distribution of possible performance estimates to use as the estimated value of the advertisement; and

publication functionality stored in the memory and executable on the processor for publishing the N candidate advertisements identified by the advertising management functionality.

16. The system of claim 15, wherein the logic configured to select a subset of N advertisements from the set of K candidate advertisements is further configured to:

rank the set of K candidate advertisements by their respective estimated expected values; and

select the subset of N advertisements as the top N ranking advertisements.

17. The system of claim 15, further configured to price the subset of N advertisements.

18. The system of claim 15, further configured to:

determine whether there is a deficiency of performance data for said at least one advertisement; and

if there is a deficiency, form the probability distribution so that it is artificially elevated, to allow said at least one advertisement to effectively compete with more mature advertisements.