Using exogenous sources for personalization of website services

1. A method, which is computer-implemented, comprising operations of:

generating a first profile for a user of a news service, wherein the first profile is based at least in part on implicit relevance feedback from the user on content presented by the news service;

obtaining a second profile for the user from a web-searching service, wherein the second profile is based on a query log with a plurality of queries by the user during a specified period of time and any search results resulting from each of the plurality of queries;

generating a first relevancy score defining how well a candidate item of content matches the first profile;

generating a second relevancy score defining how well the candidate item of content matches the second profile;

creating a combined score for the candidate item of content using score aggregation, wherein the combined score is based at least in part on similarity of the candidate item to the first profile as provided in the first relevancy score and similarity of the candidate item to the second profile as provided in the second relevancy score, and wherein similarity to the second profile measures at least similarity to each of the queries and similarity to any titles of any search results resulting from each of the queries, wherein score aggregation comprises combining the first relevancy score and the second relevancy score;

generating a final ranking for the candidate item based on the combined score, wherein the higher the combined score the more relevant the candidate item of content is to the user as reflected in the final ranking; and

presenting the candidate item of content to the user in a content stream served by the news service, based at least in part on the combined score and the final ranking.

2. The method of claim 1 , where the implicit relevance feedback includes click-throughs, click-ons, or both click-throughs and click-ons.

3. The method of claim 1 , wherein the user has expressly consented to use of the second profile when creating the score.

4. The method of claim 1 , wherein similarity is measured using cosine similarity and vectors of unigrams.

5. The method of claim 4 , further comprising an operation of:

resealing the second profile using an inverse document frequency (IDF) function.

6. The method of claim 1 , wherein the query log logs at least three months of queries and results.

7. The method of claim 1 , wherein the user is an inactive user.

8. The method of claim 1 , wherein similarity to the second profile additionally measures similarity to an abstract or snippet of a search result.

9. One or more computer-readable media which are non-transitory and which store instructions that, when executed by a processor, perform the following operations:

generate a first profile for a user of a website with multiple services including a news service and a web-searching service, wherein the first profile is based at least in part on implicit relevance feedback from the user on content presented by the news service;

obtain a second profile for the user from a web-searching service, wherein the second profile is based on a query log with a plurality of queries by the user during a specified period of time and any search results resulting from each of the plurality of queries;

generate a first relevancy score defining how well a candidate item of content matches the first profile;

generate a second relevancy score defining how well the candidate item of content matches the second profile;

create a combined score for the candidate item of content using score aggregation, wherein the combined score is based at least in part on similarity of the candidate item to the first profile as provided in the first relevancy score and similarity of the candidate item to the second profile as provided in the second relevancy score, and wherein similarity to the second profile measures at least similarity to each of the queries and similarity to any titles of any search results resulting from each of the queries, wherein score aggregation comprises combining the first relevancy score and the second relevancy score;

generate a final ranking for the candidate item based on the combined score, wherein the higher the combined score the more relevant the candidate item of content is to the user as reflected in the final ranking; and

present the candidate item of content to the user in a content stream served by the news service, based at least in part on the combined score and the final ranking.

10. The computer-readable media of claim 9 , where the implicit relevance feedback includes click-throughs, click-ons, or both click-throughs and click-ons.

11. The computer-readable media of claim 9 , wherein the user has expressly consented to use of the second profile when creating the score.

12. The computer-readable media of claim 9 , wherein similarity is measured using cosine similarity and vectors of unigrams.

13. The computer-readable media of claim 12 , further comprising an operation of:

rescale the second profile using an inverse document frequency (IDF) function.

14. The computer-readable media of claim 9 , wherein the query log logs at least three months of queries and results.

15. The computer-readable media of claim 9 , wherein the user is an inactive user.

16. The computer-readable media of claim 9 , wherein similarity to the second profile additionally measures to an abstract or snippet of a search result.

17. A method, which is computer-implemented, comprising operations of:

generating a first profile for a user of a social-networking service, wherein the first profile is based at least in part on implicit relevance feedback from the user on content presented by the social-networking service;

obtaining a second profile for the user from a web-searching service, wherein the second profile is based on a query log with a plurality of queries by the user during a specified period of time and any search results resulting from each of the plurality of queries;

generating a first relevancy score defining how well a candidate item of content matches the first profile;

generating a second relevancy score defining how well the candidate item of content matches the second profile;

creating a combined score for the candidate item of content using score aggregation, wherein the combined score is based at least in part on similarity of the candidate item to the first profile as provided in the first relevancy score and similarity of the candidate item to the second profile as provided in the second relevancy score, and wherein similarity to the second profile measures at least similarity to each of the queries and similarity to any titles of any search results resulting from each of the queries, wherein score aggregation comprises combining the first relevancy score and the second relevancy score;

generating a final ranking for the candidate item based on the combined score, wherein the higher the combined score the more relevant the candidate item of content is to the user as reflected in the final ranking; and

presenting the candidate item of content to the user in a content stream served by the social-networking service, based at least in part on the combined score and the final ranking.

18. The method of claim 17 , where the implicit relevance feedback includes likes, shares, or both likes and shares.

19. The method of claim 17 , wherein the query log logs at least three months of queries and results.

20. The method of claim 17 , wherein similarity to the second profile additionally measures similarity to an abstract or snippet of a search result.