Identifying local experts for local search

1. A method of ranking local search results based on reviews by experts, the method comprising:

receiving a query from a user via a user device;

identifying a geographic area and a category of business for the query;

selecting, with one or more processors, a plurality of candidate experts for the geographic area and the category of business defining an area-category pair, including ranking M users based on a number of reviews of businesses in the area-category pair and designating, based on the ranking, top N of the users as the plurality of candidate experts, 1<=N<M;

determining, with the one or more processors for each of the plurality of candidate experts, a respective agreement score based on differences between reviews of businesses reviewed by the user and the candidate expert;

selecting, with the one or more processors from among the plurality of candidate experts, a plurality of personalized experts in view of the determined agreement scores;

ranking, with the one or more processors, local search results responsive to the query using reviews of the local search results by the plurality of personalized experts; and

causing the ranked local search results to be displayed via the user device.

2. The method of claim 1 , wherein determining each agreement score includes determining an average difference between rating of businesses by the user and the respective candidate expert.

3. The method of claim 2 , further comprising:

calculating an expertise score for each of the plurality of candidate experts based at least in part on a number of reviews of the candidate experts;

determining, with the one or more processors, a personalization score for each of the plurality of candidate experts based a number of businesses reviewed by both the user the candidate expert and the corresponding agreement score;

wherein selecting the plurality of candidate experts includes adjusting the expertise score of a candidate expert in view of the personalization score of the candidate expert.

4. The method of claim 3 , wherein calculating the expertise score of a candidate expert includes using a text-quality score for the candidate expert, including determining the text-quality score based on one or more of (i) a n-gram correlation, or (ii) a sentence structure.

5. The method of claim 3 , wherein selecting the plurality of candidate experts includes determining whether the adjusted expertise score is above a threshold value.

6. The method of claim 1 , wherein identifying reviews for the geographic area and business category that are likely to be spam comprises:

obtaining, for at least some of the reviews, indications from users expressing whether a respective review is helpful; and

identifying reviews for the geographic area and business category that are likely to be spam in response to determining that at least some reviews are spam based on the indications.

7. The method of claim 1 , further comprising:

identifying reviews for the geographic area and business category that are shorter than a threshold length or that have less than a threshold quality score; and

removing those reviews determined to be shorter than the threshold length or to have less than the threshold quality score from a list of reviews used to identify a candidate expert.

8. The method of claim 1 , wherein identifying the plurality of candidate experts further comprises:

initializing the category of business to an initial category of business for the query;

determining that no experts are designated for the geographic area and category of business; and

iteratively decreasing specificity of the category of business until a candidate expert is designated for the modified category of business and the geographic area.

9. The method of claim 1 , further comprising:

identifying a candidate expert in the category of businesses and in the geographic area, including:

retrieving the expert from a previously formed index of experts, the index being keyed to the geographical area and the business category, and created from a classification of a plurality of reviews of a plurality of businesses in a plurality of geographic locations.

10. The method of claim 1 , comprising:

selecting an advertisement based on the query; and

sending instructions to display the advertisement.

11. A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a data processing apparatus, cause the data processing apparatus to perform operations comprising:

receiving a query from a user via a user device;

identifying a geographic area and a category of business for the query;

selecting a plurality of candidate experts for the geographic area and the category of business defining an area-category pair, including ranking M users based on a number of reviews of businesses in the area-category pair and designating, based on the ranking, top N of the users as the plurality of candidate experts, 1<=N<M,

determining, for each of the plurality of candidate experts, a respective agreement score based on differences between reviews of businesses reviewed by the user and the candidate expert;

selecting, from among the plurality of candidate experts, a plurality of personalized experts in view of the determined agreement scores;

ranking local search results responsive to the query using reviews of the local search results by the plurality of personalized experts; and

causing the ranked local search results to be displayed via the user device.

12. The machine-readable medium of claim 11 , wherein determining each agreement score includes determining an average difference between rating of businesses by the user and the respective candidate expert.

13. The machine-readable medium of claim 12 , further comprising:

calculating an expertise score for each of the plurality of candidate experts based at least in part on a number of reviews of the candidate experts;

determining, with the one or more processors, a personalization score for each of the plurality of candidate experts based a number of businesses reviewed by both the user the candidate expert and the corresponding agreement score;

wherein selecting the plurality of candidate experts includes adjusting the expertise score of a candidate expert in view of the personalization score of the candidate expert.

14. The machine-readable medium of claim 13 , wherein calculating the expertise score of a candidate expert includes using a text-quality score for the candidate expert, including determining the text-quality score based on one or more of (i) a n-gram correlation, or (ii) a sentence structure.

15. The machine-readable medium of claim 13 , wherein selecting the plurality of candidate experts includes determining whether the adjusted expertise score is above a threshold value.

16. A system, comprising:

one or more processors; and

memory storing instructions that when executed by the one or more processors cause the one or more processors to effectuate operations comprising:

receiving a query from a user via a user device,

identifying a geographic area and a category of business for the query,

selecting a plurality of candidate experts for the geographic area and the category of business defining an area-category pair, including ranking M users based on a number of reviews of businesses in the area-category pair and designating, based on the ranking, top N of the users as the plurality of candidate experts, 1<=N<M,

determining, for each of the plurality of candidate experts, a respective agreement score based on differences between reviews of businesses reviewed by the user and the candidate expert,

selecting, from among the plurality of candidate experts, a plurality of personalized experts in view of the determined agreement scores,

ranking local search results responsive to the query using reviews of the local search results by the plurality of personalized experts, and

causing the ranked local search results to be displayed via the user device.

17. The system of claim 16 , wherein determining each agreement score includes determining an average difference between rating of businesses by the user and the respective candidate expert.

18. The system of claim 17 , the operations further comprising:

calculating an expertise score for each of the plurality of candidate experts based at least in part on a number of reviews of the candidate experts;

determining a personalization score for each of the plurality of candidate experts based a number of businesses reviewed by both the user the candidate expert and the corresponding agreement score;

wherein selecting the plurality of candidate experts includes adjusting the expertise score of a candidate expert in view of the personalization score of the candidate expert.

19. The system of claim 18 , wherein calculating the expertise score of a candidate expert includes using a text-quality score for the candidate expert, including determining the text-quality score based on one or more of (i) a n-gram correlation, or (ii) a sentence structure.