Estimating effects of user interface changes on content item performance

1. A method performed by a data processing apparatus, the method comprising:

analyzing, by one or more processors, a resource provided over a network;

obtaining, based on the analysis, a first document object model representing a structure of the resource at a first time and identifying a first set of content item presentation positions;

computing, by the one or more processors, a first performance profile for the first document object model, the first performance profile specifying a plurality of performance factors, each of the plurality of performance factors based on a number of user interactions with a content item at a content item presentation position of the first set of content item presentation positions;

analyzing, by the one or more processors, the resource to obtain a second document object model representing a second structure of the resource at a second time after the first time, the second document object model identifying a second set of content item presentation positions;

computing, by the one or more processors, a similarity measure between the first document object model and the second document object model;

determining, by the one or more processors, that the similarity measure is less than a similarity threshold; and

updating, by the one or more processors based on the determination, the first performance profile for the resource to a second performance profile for the resource.

2. The method of claim 1 , wherein each performance factor specifies a normalized click-through-rate for each respective content item presentation position.

3. The method of claim 2 , further comprising:

identifying each respective content item presentation position from the first document object model for the resource.

4. The method of claim 1 , wherein computing the similarity measure comprises:

determining a first fingerprint value based on a fingerprint analysis of a structure of the first document object model;

determining a second fingerprint value based on a fingerprint analysis of a structure of the second document object model; and

comparing the first fingerprint value to the second fingerprint value.

5. The method of claim 1 , wherein updating the first performance profile for the resource to the second performance profile is based on performance of content items presented with the resource since the second time.

6. The method of claim 5 , further comprising storing the second performance profile with a reference to the second document object model.

7. The method of claim 6 , wherein storing the second performance profile comprises indexing each respective performance factor according to a content item presentation position for which the performance factor was computed.

8. The method of claim 1 , further comprising:

obtaining a third document object model representing a third structure of the resource at a third time after the second time;

computing a second similarity measure between the second document object model and the third document object model;

determining that the second similarity measure exceeds the similarity threshold; and

selecting the second performance profile for the resource.

9. The method of claim 1 , wherein the resource is a web page, the performance profile comprises a set of click-through-rate normalization factors, and the content items are advertisements.

10. The method of claim 9 , further comprising:

computing, using the performance profile, an estimated click-through-rate for an advertisement that is eligible to be presented with the web page.

11. A non-transitory computer storage medium encoded with a computer program, the program comprising instructions that when executed by data processing apparatus cause the data processing apparatus to perform operations comprising:

analyzing, by one or more processors, resource provided over a network;

obtaining, based on the analysis, a first document object model representing a structure of the resource at a first time and identifying a first set of content item presentation positions;

computing, by the one or more processors, a first performance profile for the first document object model, the first performance profile specifying a plurality of performance factors, each of the plurality of performance factors based on a number of user interactions with a content item at a content item presentation position of the first set of content item presentation positions;

analyzing, by the one or more processors, the resource to obtain a second document object model representing a second structure of the resource at a second time after the first time, the second document object model identifying a second set of content item presentation positions;

computing, by the one or more processors, a similarity measure between the first document object model and the second document object model;

determining, by the one or more processors, that the similarity measure is less than a similarity threshold; and

updating, by the one or more processors based on the determination, the first performance profile for the resource to a second performance profile for the resource.

12. The computer storage medium of claim 11 , wherein each performance factor specifies a normalized click-through-rate for each respective content item presentation position.

13. A system comprising:

a data processing apparatus including one or more processors to:

analyze, by one or more processors, resource provided over a network;

obtain, based on the analysis, a first document object model representing a structure of the resource at a first time and identifying a first set of content item presentation positions;

compute, by the one or more processors, a first performance profile for the first document object model, the first performance profile specifying a plurality of performance factors, each of the plurality of performance factors based on a number of user interactions with a content item at a content item presentation position of the first set of content item presentation positions;

analyze, by the one or more processors, the resource to obtain a second document object model representing a second structure of the resource at a second time after the first time, the second document object model identifying a second set of content item presentation positions;

compute, by the one or more processors, a similarity measure between the first document object model and the second document object model;

determine, by the one or more processors, that the similarity measure is less than a similarity threshold; and

update, by the one or more processors based on the determination, the first performance profile for the resource to a second performance profile for the resource.

14. The system of claim 13 , wherein each performance factor specifies a normalized click-through-rate for each respective content item presentation position.

15. The system of claim 13 , comprising the data processing apparatus configured to:

determine a first fingerprint value based on a fingerprint analysis of the first document object model;

determine a second fingerprint value based on a fingerprint analysis of the second document object model; and

compute the similarity measure by comparing the first fingerprint value to the second fingerprint value.

16. The system of claim 13 , comprising the data processing apparatus configured to:

compute the second performance profile based on performance of content items presented with the resource since the second time.

17. The system of claim 16 , comprising the data processing apparatus configured to:

store the second performance profile with a reference to the second document object model.

18. The system of claim 13 , comprising the data processing apparatus configured to:

obtain a third document object model representing a third structure of the resource at a third time after the second time;

compute a second similarity measure between the second document object model and the third document object model;

determine that the second similarity measure exceeds the similarity threshold; and

select the second performance profile for the resource.