Generating an implied object graph based on user behavior
An online system generates implied links between objects distributed by the online system based on a plurality of users' behaviors with respect to the objects. For a behavior of a user with respect to an object that satisfies a link criterion, the online system identifies a connection between the user and the object. Local object graphs are generated to represent the relationships between the objects connected to each user, and the local object graphs are merged into a global object graph. An adjacency matrix is extracted from the global object graph as a representation of the strength of relationships between the objects distributed by the online system. The online system may then apply graph analysis techniques to the adjacency matrix, thereby analyzing the relationships between the objects without relying upon explicit links between the objects.
1. A method for generating an implied object graph, the method comprising:
identifying a plurality of user behaviors by an online system, each user behavior comprising an interaction with an object of a plurality of objects and performed by a user of a plurality of users of the online system;
generating a plurality of local object graphs based on the user behaviors, each local object graph associated with a user of the plurality of users and comprising a subset of the objects with which the user interacted and links between objects of the subset, wherein a link between two objects of the subset is generated based on the user interaction with each of the two objects;
storing the links between objects of the subset of objects associated with each local object graph in a corresponding adjacency matrix;
for each local object graph, identifying at least one object that is common with another local object graphs of the plurality of local object graphs based on the links in the corresponding adjacency matrices associated with the plurality of local object graphs;
merging the plurality of local object graphs to generate a global object graph based on the identified common objects, the global object graph comprising the plurality of objects and links between the plurality of objects in the plurality of local object graphs; and
generating an adjacency matrix for the global object graph based on the adjacency matrices associated with the plurality of local object graphs;
receiving a request for content from the user of the online system; and
generating, by the online system, content for presentation to the user based on links between objects in the adjacency matrix for the global object graph.
2. The method of claim 1 , wherein generating, by the online system, content for presentation to the user based on links between objects in the global object graph and the adjacency matrix for the global object graph comprises:
generating a ranking of the plurality of objects based on the links between objects in the adjacency matrix for the local object graph; and
presenting at least a subset of the plurality of objects to the user in an order determined from the ranking.
3. The method of claim 2 , wherein the request for content comprises a search query and the subset of the plurality of objects in the order determined form the ranking comprises results to the search query.
4. The method of claim 1 , wherein generating, by the online system, content for presentation to the user based on links between objects in the global object graph and the adjacency matrix for the global object graph comprises:
generating a section of a digital magazine including multiple objects, each object in the section having a link to at least one other object in the section stored in the global adjacency matrix.
5. The method of claim 1 , wherein receiving the request for content from the user of the online system comprises:
receiving an interaction with an object by the user of the online system.
6. The method of claim 1 , wherein generating, by the online system, content for presentation to the user based on links between objects in the global object graph and the adjacency matrix for the global object graph comprises:
generating a ranking of the plurality of objects based on the links between objects in the adjacency matrix for the local object graph;
identifying highly ranked objects from the ranking;
identifying entities common to the identified highly ranked objects; and
presenting information identifying the entities common to the identified highly ranked objects.
7. The method of claim 1 , further comprising:
determining a weight for each of the links in the global object graph; and
ranking the objects in the online system based on the adjacency matrix associated with the global object graph and the weights.
8. The method of claim 7 , further comprising:
determining a weight for each of the links in the local object graphs, the weight determined based on the corresponding user behavior;
wherein the weight for each of the links in the global object graph is generated based on the weights of the merged local object graphs.
9. The method of claim 1 , further comprising:
determining a weight for each of the links in the local object graphs based on the corresponding user behavior; and
decaying the weight based on a time since the corresponding user behavior was identified.
10. The method of claim 1 , wherein a user behavior of the plurality of user behaviors is a user viewing the object, and wherein generating a plurality of local object graphs comprises:
responsive to the user viewing the object, identifying a connection between the user and the object.
11. The method of claim 1 , wherein generating the plurality of local objects graphs comprises:
responsive to a user of the plurality of users interacting with an object for longer than a threshold time, identifying a connection between the object and the user.
12. The method of claim 1 , wherein a user behavior of the plurality of user behaviors is an explicit feedback signal supplied by a user and corresponding to an object of the plurality of objects, and wherein generating the plurality of local object graphs comprises:
responsive to receiving the explicit feedback signal, identifying a connection between the user and the corresponding object.
13. A non-transitory computer-readable medium storing computer program instructions for generating an implied object graph, the computer program instructions comprising instructions for:
identifying a plurality of user behaviors by an online system, each user behavior comprising an interaction with an object of a plurality of objects and performed by a user of a plurality of users of the online system;
generating a plurality of local object graphs based on the user behaviors, each local object graph associated with a user of the plurality of users and comprising a subset of the objects with which the user interacted and links between objects of the subset, wherein a link between two objects of the subset is generated based on the user interaction with each of the two objects;
storing the links between objects of the subset of objects associated with each local object graph in a corresponding adjacency matrix;
for each local object graph, identifying at least one object that is common with another local object graphs of the plurality of local object graphs based on the links in the corresponding adjacency matrices associated with the plurality of local object graphs;
merging the plurality of local object graphs to generate a global object graph based on the identified common objects, the global object graph comprising the plurality of objects and links between the plurality of objects in the plurality of local object graphs;
generating an adjacency matrix for the global object graph based on the adjacency matrices associated with the plurality of local object graphs
receiving a request for content from the user of the online system; and
generating, by the online system, content for presentation to the user based on links between objects in the adjacency matrix for the global object graph.
14. The non-transitory computer-readable medium of claim 13 , wherein generating, by the online system, content for presentation to the user based on links between objects in the global object graph and the adjacency matrix for the global object graph comprises:
generating a ranking of the plurality of objects based on the links between objects in the adjacency matrix for the local object graph; and
presenting at least a subset of the plurality of objects to the user in an order determined from the ranking.
15. The non-transitory computer-readable medium of claim 14 , wherein the request for content comprises a search query and the subset of the plurality of objects in the order determined form the ranking comprises results to the search query.
16. The non-transitory computer-readable medium of claim 13 , wherein generating, by the online system, content for presentation to the user based on links between objects in the global object graph and the adjacency matrix for the global object graph comprises:
generating a section of a digital magazine including multiple objects, each object in the section having a link to at least one other object in the section stored in the global adjacency matrix.
17. The non-transitory computer-readable medium of claim 13 , wherein receiving the request for content from the user of the online system comprises:
receiving an interaction with an object by the user of the online system.
18. The non-transitory computer-readable medium of claim 13 , wherein generating, by the online system, content for presentation to the user based on links between objects in the global object graph and the adjacency matrix for the global object graph comprises:
generating a ranking of the plurality of objects based on the links between objects in the adjacency matrix for the local object graph;
identifying highly ranked objects from the ranking;
identifying entities common to the identified highly ranked objects; and
presenting information identifying the entities common to the identified highly ranked objects.
19. The non-transitory computer-readable medium of claim 13 , the instructions further comprising instructions for:
determining a weight for each of the links in the global object graph,
ranking the objects in the online system based on the adjacency matrix associated with the global object graph and the weights.
20. The non-transitory computer-readable medium of claim 13 , the instructions further comprising instructions for:
determining a weight for each of the links in the local object graphs based on a type of the corresponding user behavior.
21. The non-transitory computer-readable medium of claim 13 , wherein a user behavior of the plurality of user behaviors is a user viewing the object, and wherein generating a plurality of local object graphs comprises:
responsive to the user viewing the object, identifying a connection between the user and the object.
22. The non-transitory computer-readable medium of claim 13 , wherein generating the plurality of local objects graphs comprises:
responsive to a user of the plurality of users interacting with an object for longer than a threshold time, identifying a connection between the object and the user.