INFERRING USER INTERESTS

• US 20120054205A1
• Filed: 11/07/2011
• Published: 03/01/2012
• Est. Priority Date: 05/01/2007
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Baluja, Shumeet, YAGNIK, Jay, JING, Yushi, SIVAKUMAR, Dandapani
• Granted Patent: US 8,473,500 B2
• Time in Patent Office: Days
• US Class Current: 707/749
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G06Q 30/0274 Split fees

Abstract

The subject matter of this specification can be embodied in, among other things, a method that includes determining, for a portion of users of a social network, label values each comprising an inferred interest level of a user in a subject indicated by a label, associating a first user with one or more second users based on one or more relationships specified by the first user, and outputting a first label value for the first user based on one or more second label values of the one or more second users.

Claims

1. A computer-implemented method comprising:

accessing, by a computer system, at least a portion of a graph that includes i) nodes that are associated with users of a social network and ii) edges that connect nodes of the graph;

selecting, by the computer system, a node from the accessed portion of the graph, wherein the node is associated with a user of the social network;

identifying, from the accessed portion of the graph, one or more neighboring nodes that are connected to the selected node by one or more of the edges, wherein the one or more neighboring nodes are associated with one or more other users of the social network;

determining, for the selected node, a plurality of interest values that indicate levels of interest of the user in a plurality of subjects, wherein the plurality of interest values are determined based on a plurality of neighbor interest values that are i) associated with the one or more neighboring nodes and ii) that indicate levels of interest of the one or more other users in the plurality of subjects;

adjusting each of at least a portion of the plurality of interest values to account for a relative strength of the user'"'"'s level of interest in each of at least a portion of the plurality of subjects, wherein each of the at least a portion of the plurality of interest values are adjusted based, at least in part, on a combination of the plurality of interest values; and

providing the adjusted plurality of interest values in association with the selected node of the graph.

2. The method of claim 1, wherein adjusting each of the at least a portion of the plurality of interest values comprises normalizing each of the at least a portion of the plurality of interest values based on the combination of the plurality of interest values.

3. The method of claim 2, wherein normalizing each of the at least a portion of the plurality of interest values based on the combination of the plurality of interest values comprises normalizing each of the plurality of interest values based on a sum of the plurality of interest values.

4. The method of claim 2, wherein the selecting, identifying, determining, adjusting, and providing steps are performed with respect to each node in the accessed portion of the graph.

5. The method of claim 1, wherein the combination of the plurality of interest values comprises a sum of the plurality of interest values.

6. The method of claim 5, wherein each of the plurality of interest values is determined based on one of:

a current interest value, a combination of one or more neighbor interest values, or a combination of one or more neighbor interest values and a current interest value;

where the current interest value is associated with a current state of the selected node and indicates a level of interest of the user in a subject; and

where the one or more neighbor interest values are i) associated with the one or more neighboring nodes and ii) indicate levels of interest of the one or more other users in the subject.

7. The method of claim 6, wherein the combinations are sums.

8. The method of claim 1, wherein determining the plurality of values includes, at least:

determining, for the selected node, a first interest value that indicates a first level of interest of the user in a first subject, wherein the first interest value is determined based on a first combination of one or more first neighbor interest values, wherein the first neighbor interest values are i) associated with the one or more neighboring nodes and ii) indicate levels of interest of the one or more other users in the first subject; and

determining, for the selected node, a second interest value that indicates a second level of interest of the user in a second subject, wherein the second interest value is determined based on a second combination of one or more second neighbor interest values, wherein the one or more second neighbor interest values are i) associated with the one or more neighboring nodes and ii) that indicate levels of interest of the one or more other users in the second subject;

wherein adjusting each of the at least a portion of the plurality of interest values includes, at least, adjusting the first interest value based on the combination of the plurality of interest values that includes both the first interest value and the second interest value.

9. The method of claim 1, wherein at least a portion of the edges are associated with weights that indicate strength of corresponding relationships between nodes of the graph, and wherein the plurality of interest values are further determined based on at least a portion of the weights in combination with at least a portion of the plurality of neighbor interest values.

10. The method of claim 9, wherein determining the plurality of interest values includes weighting, using one or more of the weights that are associated with particular edges connecting the selected node to particular neighboring nodes, particular neighbor interest values for the particular neighboring nodes.

11. The method of claim 10, wherein weighting comprises multiplying the particular neighbor interest values by corresponding weights.

12. The method of claim 1, wherein at least the determining, adjusting and providing steps are performed iteratively.

13. The method of claim 12, wherein the iterated steps are stopped when each of the plurality of interest values are within a predetermined threshold of the interest values from the previous iteration.

14. The method of claim 12, wherein when performing the selecting, the identifying, the determining, and the adjusting for a second node that is connected to the selected node by one or more of the edges, a second plurality of interest values are determined for the second node using the adjusted plurality of interest values for the selected node.

15. A system comprising:

a computer system;

a data structure generator of the computer system to access at least a portion of a graph that includes i) nodes that are associated with users of a social network and ii) edges that connect nodes of the graph;

an inferred label generator of the computer system to;

select a node from the accessed portion of the graph, wherein the node is associated with a user of the social network;

identify, from the accessed portion of the graph, one or more neighboring nodes that are connected to the selected node by one or more of the edges, wherein the one or more neighboring nodes are associated with one or more other users of the social network;

determine, for the selected node, a plurality of interest values that indicate levels of interest of the user in a plurality of subjects, wherein the plurality of interest values are determined based on a plurality of neighbor interest values that are i) associated with the one or more neighboring nodes and ii) that indicate levels of interest of the one or more other users in the plurality of subjects; and

adjust each of at least a portion of the plurality of interest values to account for a relative strength of the user'"'"'s level of interest in each of at least a portion of the plurality of subjects, wherein each of the at least a portion of the plurality of interest values are adjusted based, at least in part, on a combination of the plurality of interest values; and

an interface of the computer system to provide the adjusted plurality of interest values in association with the selected node of the graph.

16. The system of claim 15, wherein adjusting each of the at least a portion of the plurality of interest values comprises normalizing each of the at least a portion of the plurality of interest values based on the combination of the plurality of interest values.

17. The system of claim 16, wherein normalizing each of the at least a portion of the plurality of interest values based on the combination of the plurality of interest values comprises normalizing each of the plurality of interest values based on a sum of the plurality of interest values.

18. The system of claim 16, wherein the selecting, identifying, determining, adjusting, and providing steps are performed with respect to each node in the accessed portion of the graph.

19. The system of claim 15, wherein the combination of the plurality of interest values comprises a sum of the plurality of interest values.

20. The system of claim 19, wherein each of the plurality of interest values is determined based on one of:

a current interest value, a combination of one or more neighbor interest values, or a combination of one or more neighbor interest values and a current interest value;

where the current interest value is associated with a current state of the selected node and indicates a level of interest of the user in a subject; and

where the one or more neighbor interest values are i) associated with the one or more neighboring nodes and ii) indicate levels of interest of the one or more other users in the subject.

21. The system of claim 20, wherein the combinations are sums.

22. The system of claim 15, wherein determining the plurality of values includes, at least:

determining, for the selected node, a first interest value that indicates a first level of interest of the user in a first subject, wherein the first interest value is determined based on a first combination of one or more first neighbor interest values, wherein the first neighbor interest values are i) associated with the one or more neighboring nodes and ii) indicate levels of interest of the one or more other users in the first subject; and

determining, for the selected node, a second interest value that indicates a second level of interest of the user in a second subject, wherein the second interest value is determined based on a second combination of one or more second neighbor interest values, wherein the one or more second neighbor interest values are i) associated with the one or more neighboring nodes and ii) that indicate levels of interest of the one or more other users in the second subject;

wherein adjusting each of the at least a portion of the plurality of interest values includes, at least, adjusting the first interest value based on the combination of the plurality of interest values that includes both the first interest value and the second interest value.

23. The system of claim 15, wherein at least a portion of the edges are associated with weights that indicate strength of corresponding relationships between nodes of the graph, and wherein the plurality of interest values are further determined based on at least a portion of the weights in combination with at least a portion of the plurality of neighbor interest values.

24. The system of claim 21, wherein determining the plurality of interest values includes weighting, using one or more of the weights that are associated with particular edges connecting the selected node to particular neighboring nodes, particular neighbor interest values for the particular neighboring nodes.

25. The system of claim 24, wherein weighting comprises multiplying the particular neighbor interest values by corresponding weights.

26. The system of claim 15, wherein the inferred label generator iteratively performs at least the determining, adjusting, and providing steps.

27. The system of claim 24, wherein the iterated steps are stopped when each of the plurality of interest values are within a predetermined threshold of the interest values from the previous iteration.

28. The system of claim 24, wherein, when the inferred label generator performs the selecting, the identifying, the determining, and the adjusting for a second node that is connected to the selected node by one or more of the edges, the inferred label generator uses the adjusted plurality of interest values for the selected node to determine a second plurality of interest values for the second node.

29. A computer-implemented method comprising:

accessing, by a computer system, at least a portion of a graph that includes i) nodes that are associated with users of a social network and ii) edges that connect nodes of the graph;

selecting, by the computer system, a first node from the accessed portion of the graph, wherein the first node is associated with a first user of the social network;

identifying, from the accessed portion of the graph, a second node that is associated with a second user of the social network and that is connected to the first node by a first edge, and a third node that is associated with a third user of the social network and that is connected to the first node by a second edge;

determining, for the first node, a first interest value and a second interest value that indicate a level of interest of the first user in a first subject and a level of interest of the first user in a second subject, respectively, wherein the first and second interest values are determined based, at least in part, on i) a third interest value and a fourth interest value that indicate a level of interest of the second user in the first subject and a level of interest of the first user in a second subject, respectively, and ii) a fifth interest value and a sixth interest value that indicate a level of interest of the third user in the first subject and a level of interest of the third user in the second subject, respectively;

adjusting the first interest value and the second interest value for the first node to account for a relative strength of the first user'"'"'s level of interest in at least the first subject and the second subject, wherein the adjusting includes i) normalizing the first interest value based on a combination of, at least, the first interest value and the second interest value, and ii) normalizing the second interest value based on the combination of, at least, the first interest value and the second interest value; and

providing the adjusted first interest value and the adjusted second interest value in association with the first node of the graph.