Cross-Validating Places on Online Social Networks

1 . A method comprising, by one or more computer systems of an online social network:

accessing a place-entities graph comprising a plurality of place-entity nodes, each place-entity node representing a place-entity corresponding to a particular geographic location, each edge between two nodes establishing a single degree of separation between them;

identifying a place-entity cluster within the place-entities graph, wherein the place-entity cluster comprises a plurality of place-entity nodes corresponding to a plurality of place-entities, respectively, each place-entity corresponding to the same geographic location;

accessing a plurality of embeddings representing the plurality of place-entities corresponding to the place-entity cluster, respectively, wherein each embedding is a point in a d-dimensional embedding space;

calculating, using a machine-learning model, a cluster-quality score of the place-entity cluster based on the plurality of embeddings representing the place-entities corresponding to the place-entity cluster, wherein the cluster-quality score represents a probability that the place-entities corresponding to the place-entity cluster correspond to a valid geographic location; and

identifying the place-entities corresponding to the place-entity cluster as corresponding to an invalid geographic location based on a determining that the cluster-quality score is less than a threshold cluster-quality score.

2 . The method of claim 1 , wherein each place-entity cluster comprises place-entity nodes having duplication-values with respect to a canonical place-entity node in the place-entity cluster above a threshold duplication-value.

3 . The method of claim 1 further comprising, for each place-entity of the plurality of place-entities, generating the embedding representing the place-entity.

4 . The method of claim 3 , wherein, for each place-entity, the embedding representing the place-entity was generated based at least in part on one or more of a source of information associated with the place-entity, a date of information associated with the place-entity, an accuracy of attributes of the place-entity, a number of photos associated with the place-entity, an amount of content associated with the place-entity, a recency-value associated with the content or photos, or a number of social signals associated with the place-entity.

5 . The method of claim 4 , wherein the source of information associated with at least one place-entity is a user of the online social network or a third-party system.

6 . The method of claim 1 further comprising:

receiving, from a client system associated with a user, a query comprising one or more n-grams;

identifying one or more place-entities matching at least a portion of the query, wherein the identified place-entities comprise at least one place-entity corresponding to the place-entity cluster;

calculating, for each identified place-entity, a relevance score;

ranking the identified place-entities based at least in part on their respective relevance scores, wherein each identified place-entity corresponding to the place-entity cluster is ranked lower than each identified place-entity not corresponding to the place-entity cluster; and

sending, to the client system responsive to the query, instructions for presenting a search-results page, the search-results page comprising one or more search results corresponding to the one or more identified place-entities, respectively, wherein the search-results page displays the search results ordered by the rank of the corresponding identified place-entities.

7 . The method of claim 1 further comprising:

receiving, from a client system associated with a user, a query comprising one or more n-grams;

identifying one or more place-entities matching at least a portion of the query, wherein the identified place-entities comprise at least one place-entity corresponding to the place-entity cluster; and

sending, to the client system responsive to the query, instructions for presenting a search-results page, the search-results page comprising one or more search results corresponding to the one or more identified place-entities, respectively, wherein each search result corresponds to an identified place-entity that does not correspond to the place-entity cluster.

8 . The method of claim 1 , further comprising calculating, for each place-entity, an entity-quality score of the place-entity.

9 . The method of claim 8 , wherein each embedding representing a place-entity was generated based at least in part on whether the place-entity has an entity-quality score less than a threshold entity-quality score.

10 . One or more computer-readable non-transitory storage media embodying software that is operable when executed to:

access a place-entities graph comprising a plurality of place-entity nodes, each place-entity node representing a place-entity corresponding to a particular geographic location, each edge between two nodes establishing a single degree of separation between them;

identify a place-entity cluster within the place-entities graph, wherein the place-entity cluster comprises a plurality of place-entity nodes corresponding to a plurality of place-entities, respectively, each place-entity corresponding to the same geographic location;

access a plurality of embeddings representing the plurality of place-entities corresponding to the place-entity cluster, respectively, wherein each embedding is a point in a d-dimensional embedding space;

calculate, using a machine-learning model, a cluster-quality score of the place-entity cluster based on the plurality of embeddings representing the place-entities corresponding to the place-entity cluster, wherein the cluster-quality score represents a probability that the place-entities corresponding to the place-entity cluster correspond to a valid geographic location; and

identify the place-entities corresponding to the place-entity cluster as corresponding to an invalid geographic location based on a determining that the cluster-quality score is less than a threshold cluster-quality score.

11 . The media of claim 10 , wherein each place-entity cluster comprises place-entity nodes having duplication-values with respect to a canonical place-entity node in the place-entity cluster above a threshold duplication-value.

12 . The media of claim 10 , wherein the software is further operable when executed to, for each place-entity of the plurality of place-entities, generate the embedding representing the place-entity.

13 . The media of claim 12 , wherein, for each place-entity, the embedding representing the place-entity was generated based at least in part on one or more of a source of information associated with the place-entity, a date of information associated with the place-entity, an accuracy of attributes of the place-entity, a number of photos associated with the place-entity, an amount of content associated with the place-entity, a recency-value associated with the content or photos, or a number of social signals associated with the place-entity.

14 . The media of claim 13 , wherein the source of information associated with at least one place-entity is a user of the online social network or a third-party system.

15 . The media of claim 10 , wherein the software is further operable when executed to:

receive, from a client system associated with a user, a query comprising one or more n-grams;

identify one or more place-entities matching at least a portion of the query, wherein the identified place-entities comprise at least one place-entity corresponding to the place-entity cluster;

calculate, for each identified place-entity, a relevance score;

rank the identified place-entities based at least in part on their respective relevance scores, wherein each identified place-entity corresponding to the place-entity cluster is ranked lower than each identified place-entity not corresponding to the place-entity cluster; and

send, to the client system responsive to the query, instructions for presenting a search-results page, the search-results page comprising one or more search results corresponding to the one or more identified place-entities, respectively, wherein the search-results page displays the search results ordered by the rank of the corresponding identified place-entities.

16 . The media of claim 10 , wherein the software is further operable when executed to:

receive, from a client system associated with a user, a query comprising one or more n-grams;

identify one or more place-entities matching at least a portion of the query, wherein the identified place-entities comprise at least one place-entity corresponding to the place-entity cluster; and

send, to the client system responsive to the query, instructions for presenting a search-results page, the search-results page comprising one or more search results corresponding to the one or more identified place-entities, respectively, wherein each search result corresponds to an identified place-entity that does not correspond to the place-entity cluster.

17 . The media of claim 10 , wherein the software is further operable when executed to calculate, for each place-entity, an entity-quality score of the place-entity.

18 . The media of claim 17 , wherein each embedding representing a place-entity was generated based at least in part on whether the place-entity has an entity-quality score less than a threshold entity-quality score.

19 . A system comprising: one or more processors; and a non-transitory memory coupled to the processors comprising instructions executable by the processors, the processors operable when executing the instructions to:

access a place-entities graph comprising a plurality of place-entity nodes, each place-entity node representing a place-entity corresponding to a particular geographic location, each edge between two nodes establishing a single degree of separation between them;

identify a place-entity cluster within the place-entities graph, wherein the place-entity cluster comprises a plurality of place-entity nodes corresponding to a plurality of place-entities, respectively, each place-entity corresponding to the same geographic location;

access a plurality of embeddings representing the plurality of place-entities corresponding to the place-entity cluster, respectively, wherein each embedding is a point in a d-dimensional embedding space;

calculate, using a machine-learning model, a cluster-quality score of the place-entity cluster based on the plurality of embeddings representing the place-entities corresponding to the place-entity cluster, wherein the cluster-quality score represents a probability that the place-entities corresponding to the place-entity cluster correspond to a valid geographic location; and

identify the place-entities corresponding to the place-entity cluster as corresponding to an invalid geographic location based on a determining that the cluster-quality score is less than a threshold cluster-quality score.

20 . The system of claim 19 , wherein each place-entity cluster comprises place-entity nodes having duplication-values with respect to a canonical place-entity node in the place-entity cluster above a threshold duplication-value.