Re-ranking search results using blended learning models

1. A machine-readable medium having executable instructions to cause one or more processing units to perform a method to re-rank a plurality of search results, the method comprising:

receiving, on a client device, a plurality of crowd-source intra-domain ranking models and a crowd-sourced inter-domain ranking model, wherein each of the crowd-sourced ranking models is based on at least on-device interactions of a plurality of users interacting with a plurality of other devices;

generating, on the client device, a re-ranking model that includes a plurality of intra-domain models and an inter-domain model, wherein the plurality of intra-domain models are generated based on at least the plurality of crowd-sourced intra-domain ranking models and on-device sensitive data that remains on the client device and is not sent to a server, the inter-domain model indicates which of a plurality of search domains are to be ranked higher than others in the plurality of search domains and is generated based on at least the on-device sensitive data and the crowd-sourced inter-domain ranking model, and wherein each of the plurality of search domains corresponds to one of the plurality of intra-domain models;

receiving, on the client device, a plurality of search results generated over the plurality of search domains, wherein the plurality of search results is ranked according to a first ranking, and the plurality of search results are generated based on at least a search query;

re-ranking the plurality of search results, by the client device, using the re-ranking model to generate a second ranking of the plurality of search results; and

presenting the plurality of search results using the second ranking.

2. The machine-readable medium of claim 1 , wherein the plurality of search domains includes a plurality of on-device search domains.

3. The machine-readable medium of claim 2 , wherein each of the plurality of on-device search domains is selected from the group consisting of text messages, emails, contacts, calendar, contacts, music, movies, photos, application states, and installed applications.

4. The machine-readable medium of claim 1 , wherein the intra-domain model is further generated from private information of device user stored on the client device.

5. The machine-readable medium of claim 1 , wherein the re-ranking of the plurality of results further comprises:

re-ranking a domain subset of the plurality of results in one of the plurality of search domains using the corresponding intra-domain model.

6. The machine-readable medium of claim 1 , wherein the plurality of search domains includes a plurality of off-device search domains.

7. The machine-readable medium of claim 6 , wherein each of the plurality of off-device search domains is selected from the group consisting of maps search domain, media store search domain, online encyclopedia search domain, and sites search domain.

8. The machine-readable medium of claim 1 , wherein the re-ranking of the plurality of results further comprises: re-ranking the plurality of search domains using the inter-domain model.

9. The machine-readable medium of claim 1 , wherein the inter-domain model is further generated from private information of device user stored on the client device.

10. The machine-readable medium of claim 1 , wherein the re-ranking model further includes a policy model.

11. A method to re-rank a plurality of search results, the method comprising:

receiving, on a client device, a plurality of crowd-source intra-domain ranking models and a crowd-sourced inter-domain ranking model, wherein each of the crowd-sourced ranking models is based on at least on-device interactions of a plurality of users interacting with a plurality of other devices;

generating, on the client device, a re-ranking model that includes a plurality of intra-domain models and an inter-domain model, wherein the plurality of intra-domain models are generated based on at least the plurality of crowd-sourced intra-domain ranking models and on-device sensitive data that remains on the client device and is not sent to a server, the inter-domain model indicates which of a plurality of search domains are to be ranked higher than others in the plurality of search domains and is generated based on at least the on-device sensitive data and the crowd-sourced inter-domain ranking model, and wherein each of the plurality of search domains corresponds to one of the plurality of intra-domain models;

receiving, on the client device, a plurality of search results generated over the plurality of search domains, wherein the plurality of search results is ranked according to a first ranking, and the plurality of search results are generated based on at least a search query;

re-ranking the plurality of search results, by the client device, using the re-ranking model to generate a second ranking of the plurality of search results; and

presenting the plurality of search results using the second ranking.

12. The method of claim 11 , wherein the plurality of search domains includes a plurality of on-device search domains.

13. The method of claim 12 , wherein each of the plurality of on-device search domains is selected from the group consisting of text messages, emails, contacts, calendar, contacts, music, movies, photos, application states, and installed applications.

14. The method of claim 11 , wherein the intra-domain model is further generated from private information of device user stored on the client device.

15. The method of claim 11 , wherein the re-ranking of the plurality of results further comprises:

re-ranking a domain subset of the plurality of results in one of the plurality of search domains using the corresponding intra-domain model.

16. The method of claim 11 , wherein the plurality of search domains includes a plurality of off-device search domains.

17. The method of claim 16 , wherein each of the plurality of off-device search domains is selected from the group consisting of maps search domain, media store search domain, online encyclopedia search domain, and sites search domain.

18. A device to re-rank a plurality of search results, the device comprising:

a processor;

a memory coupled to the processor though a bus; and

a process executed from the memory by the processor causes the processor to

receive a plurality of crowd-source intra-domain ranking models and a crowd-sourced inter-domain ranking model, wherein each of the crowd-sourced ranking models is based on at least on-device interactions of a plurality of users interacting with a plurality of other devices;

generate a re-ranking model that includes a plurality of intra-domain models and an inter-domain model, wherein the plurality of intra-domain models are generated based on at least the plurality of crowd-sourced intra-domain ranking models and on-device sensitive data that remains on the device and is not sent to a server, the inter-domain model indicates which of a plurality of search domains are to be ranked higher than others in the plurality of search domains and is generated based on at least the on-device sensitive data and the crowd-sourced inter-domain ranking model, and wherein each of the plurality of search domains corresponds to one of the plurality of intra-domain models,

receive a plurality of search results generated over the plurality of search domains, wherein the plurality of search results is ranked according to a first ranking, and the plurality of search results are generated based on at least a search query,

re-rank the plurality of search results using the re-ranking model to generate a second ranking of the plurality of search results, and

present the plurality of search results using the second ranking.