Detecting a page for a real-world entity, an imposter of a real-world entity, or a non-real-world entity that complies with or violates a policy of an online system
An online system maintains pages and accesses a graph of nodes representing the pages. Each node is labeled to indicate that a corresponding page is for a real-world entity, an imposter of the real-world entity, or a derived entity complying with or violating a policy. The online system retrieves machine-learning models, each of which is trained based on labels for a set of the nodes and features of corresponding pages. A first model predicts whether a page is for a derived entity based on features of the page. Responsive to predicting the page is not for a derived entity, a second model predicts whether the page is for a real-world entity or an imposter based on features of the page. Responsive to predicting the page is for a derived entity, a third model predicts whether the derived entity complies with or violates the policy based on features of the page.
1. A method comprising:
maintaining, at an online system, a plurality of pages for one or more of: a set of real-world entities, a set of imposters of the set of real-world entities, and a set of derived entities; wherein the set of derived entities is a set of non-real-world entities related to the set of real-world entities;
accessing a graph comprising a plurality of nodes representing the plurality of pages, each of the plurality of nodes having a label indicating that a corresponding page is for a real-world entity, an imposter of the real-world entity, a derived entity that complies with a policy of the online system, or a derived entity that violates the policy;
retrieving a plurality of machine-learning models, each of the plurality of machine-learning models trained based at least in part on a set of labels for a set of the plurality of nodes and a set of features of each of a set of pages corresponding to the set of the plurality of nodes;
using a first machine-learning model of the plurality of machine-learning models to predict whether a page maintained in the online system is for a derived entity based at least in part on one or more features of the page;
responsive to predicting that the page is not for a derived entity, using a second machine-learning model of the plurality of machine-learning models to predict whether the page is for a real-world entity or an imposter of a real-world entity based at least in part on the one or more features of the page; and
responsive to predicting that the page is for a derived entity, using a third machine-learning model of the plurality of machine-learning models to predict whether the page is for a derived entity that complies with the policy or a derived entity that violates the policy based at least in part on the one or more features of the page.
2. The method of claim 1 , wherein the policy prohibits impersonating an entity.
3. The method of claim 1 , wherein the policy prohibits creating a page that is misleading.
4. The method of claim 1 , wherein the set of features of each of the set of pages comprises one or more of: a title of each of the set of pages, a name associated with each of the set of pages, a topic of each of the set of pages, a description of each of the set of pages, an image associated with each of the set of pages, information describing user engagement with each of the set of pages, a date that each of the set of pages was created, a fan count for each of the set of pages, a ratio of fans to followers of each of the set of pages, a ratio of content posted to each of the set of pages that is re-shared, information describing a violation of the policy by each of the set of pages, and content posted to each of the set of pages by an administrator of each of the set of pages.
5. The method of claim 4 , wherein the information describing user engagement with each of the set of pages comprises one or more selected from the group consisting of:
content posted by one or more users of the online system to each of the set of pages, a set of features of a user of the online system who performed an interaction with each of the set of pages, a number of content items posted to each of the set of pages, a number of comments on each of the set of pages, a number of times that each of the set of pages was shared, a number of times that one or more users of the online system expressed a preference for each of the set of pages, and a number of users of the online system who have established a connection to each of the set of pages.
6. The method of claim 4 , wherein the content posted to each of the set of pages by the administrator of each of the set of pages comprises one or more of: text included among the content and a set of features of an image included among the content.
7. The method of claim 1 , further comprising:
training the first machine-learning model based at least in part on the label for each of the plurality of nodes and the set of features of each of the plurality of pages;
training the second machine-learning model based at least in part on the label for each of a first set of nodes and the set of features of each of a first set of pages corresponding to the first set of nodes, the label for each of the first set of nodes indicating that the corresponding page is for a real-world entity or an imposter of the real-world entity; and
training the third machine-learning model based at least in part on the label for each of a second set of nodes and the set of features of each of a second set of pages corresponding to the second set of nodes, the label for each of the second set of nodes indicating that the corresponding page is for a derived entity that complies with the policy or a derived entity that violates the policy.
8. The method of claim 1 , further comprising:
generating the graph comprising the plurality of nodes based at least in part on a measure of similarity between a page that has been verified to be for a real-world entity and an additional plurality of pages maintained in the online system.
9. The method of claim 8 , wherein generating the graph comprising the plurality of nodes comprises:
generating one or more embeddings corresponding to the page that has been verified to be for the real-world entity based at least in part on a set of features of the page that has been verified to be for the real-world entity;
generating one or more additional embeddings corresponding to each of the additional plurality of pages based at least in part on a set of features of each of the additional plurality of pages;
identifying a plurality of embeddings corresponding to the plurality of pages, wherein the plurality of embeddings is within a threshold distance of the one or more embeddings corresponding to the page that has been verified to be for the real-world entity;
generating the plurality of nodes representing the plurality of pages;
assigning the label to each of the plurality of nodes based at least in part on a set of features of each of the plurality of pages; and
generating one or more edges connecting the plurality of nodes based at least in part on the label assigned to each of the plurality of nodes.
10. The method of claim 9 , wherein the label is manually assigned to each of the plurality of nodes.
11. A computer program product comprising a non-transitory computer readable storage medium having instructions encoded thereon that, when executed by a processor, cause the processor to:
maintain, at an online system, a plurality of pages for one or more of: a set of real-world entities, a set of imposters of the set of real-world entities, and a set of derived entities; wherein the set of derived entities is a set of non-real-world entities related to the set of real-world entities;
access a graph comprising a plurality of nodes representing the plurality of pages, each of the plurality of nodes having a label indicating that a corresponding page is for a real-world entity, an imposter of the real-world entity, a derived entity that complies with a policy of the online system, or a derived entity that violates the policy;
retrieve a plurality of machine-learning models, each of the plurality of machine-learning models trained based at least in part on a set of labels for a set of the plurality of nodes and a set of features of each of a set of pages corresponding to the set of the plurality of nodes;
use a first machine-learning model of the plurality of machine-learning models to predict whether a page maintained in the online system is for a derived entity based at least in part on one or more features of the page;
responsive to predicting that the page is not for a derived entity, use a second machine-learning model of the plurality of machine-learning models to predict whether the page is for a real-world entity or an imposter of a real-world entity based at least in part on the one or more features of the page; and
responsive to predicting that the page is for a derived entity, use a third machine-learning model of the plurality of machine-learning models to predict whether the page is for a derived entity that complies with the policy or a derived entity that violates the policy based at least in part on the one or more features of the page.
12. The computer program product of claim 11 , wherein the policy prohibits impersonating an entity.
13. The computer program product of claim 11 , wherein the policy prohibits creating a page that is misleading.
14. The computer program product of claim 11 , wherein the set of features of each of the set of pages comprises one or more of: a title of each of the set of pages, a name associated with each of the set of pages, a topic of each of the set of pages, a description of each of the set of pages, an image associated with each of the set of pages, information describing user engagement with each of the set of pages, a date that each of the set of pages was created, a fan count for each of the set of pages, a ratio of fans to followers of each of the set of pages, a ratio of content posted to each of the set of pages that is re-shared, information describing a violation of the policy by each of the set of pages, and content posted to each of the set of pages by an administrator of each of the set of pages.
15. The computer program product of claim 14 , wherein the information describing user engagement with each of the set of pages comprises one or more selected from the group consisting of: content posted by one or more users of the online system to each of the set of pages, a set of features of a user of the online system who performed an interaction with each of the set of pages, a number of content items posted to each of the set of pages, a number of comments on each of the set of pages, a number of times that each of the set of pages was shared, a number of times that one or more users of the online system expressed a preference for each of the set of pages, and a number of users of the online system who have established a connection to each of the set of pages.
16. The computer program product of claim 14 , wherein the content posted to each of the set of pages by the administrator of each of the set of pages comprises one or more of: text included among the content and a set of features of an image included among the content.
17. The computer program product of claim 11 , wherein the computer-readable storage medium further has instructions encoded thereon that, when executed by the processor, cause the processor to:
train the first machine-learning model based at least in part on the label for each of the plurality of nodes and the set of features of each of the plurality of pages;
train the second machine-learning model based at least in part on the label for each of a first set of nodes and the set of features of each of a first set of pages corresponding to the first set of nodes, the label for each of the first set of nodes indicating that the corresponding page is for a real-world entity or an imposter of the real-world entity; and
train the third machine-learning model based at least in part on the label for each of a second set of nodes and the set of features of each of a second set of pages corresponding to the second set of nodes, the label for each of the second set of nodes indicating that the corresponding page is for a derived entity that complies with the policy or a derived entity that violates the policy.
18. The computer program product of claim 11 , wherein the computer-readable storage medium further has instructions encoded thereon that, when executed by the processor, cause the processor to:
generate the graph comprising the plurality of nodes based at least in part on a measure of similarity between a page that has been verified to be for a real-world entity and an additional plurality of pages maintained in the online system.
19. The computer program product of claim 18 , wherein generate the graph comprising the plurality of nodes comprises:
generate one or more embeddings corresponding to the page that has been verified to be for the real-world entity based at least in part on a set of features of the page that has been verified to be for the real-world entity;
generate one or more additional embeddings corresponding to each of the additional plurality of pages based at least in part on a set of features of each of the additional plurality of pages;
identify a plurality of embeddings corresponding to the plurality of pages, wherein the plurality of embeddings is within a threshold distance of the one or more embeddings corresponding to the page that has been verified to be for the real-world entity;
generate the plurality of nodes representing the plurality of pages;
assign the label to each of the plurality of nodes based at least in part on a set of features of each of the plurality of pages; and
generate one or more edges connecting the plurality of nodes based at least in part on the label assigned to each of the plurality of nodes.
20. The computer program product of claim 19 , wherein the label is manually assigned to each of the plurality of nodes.