Simulated annealing in recommendation systems

• US 8,930,392 B1
• Filed: 06/05/2012
• Issued: 01/06/2015
• Est. Priority Date: 06/05/2012
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Leichter, Jerrold
• Primary Examiner(s): Peach, Polina
• Application Number:
  US13/489,338
• Time in Patent Office: 945 Days
• Field of Search: 707767-768
• US Class Current: 707/767
• CPC Class Codes: G06F 16/3322 using system suggestions G06F16/3325 takes precedence
• CPC Class Codes: G06F 16/3326 using relevance feedback from the user, e.g. relevance feedback on documents, documents sets, document terms or passages
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation

Abstract

A system and method for generating suggestions to a user using simulated annealing is disclosed. The system includes a user activity processing unit, a recommendation engine, a ranking engine and a simulated annealing engine. The user activity processing unit receives user activities and generates user interests. The simulated annealing engine generates a temperature to control how far to stray from highly relevant suggestions. The recommendation engine generates a query for content and generates at least one suggestion for the user from received content. The ranking engine ranks the at least one suggestion based on a recency of user interests, frequency of actions associated with the user interests and social affinity of the user to other users that generated any of the at least one suggestions.

Claims

1. A method comprising:

identifying, by one or more processors, a plurality of user interests from user activity associated with a user;

applying a weight to each of the user interests based on a frequency of user activity related to each of the user interests;

generating, by one or more processors, groups on a social network for the user, at least a first group from the set of the groups including a user interest;

generating, by the one or more processors, a query for content for the first group that includes the user interest;

receiving, by the one or more processors, the content for the first group;

generating, by the one or more processors, a set of ranked suggestions from the content for the first group based on relevancy to the user interest, the weight of the user interest and explicit feedback from the user, the explicit feedback including a first setting for how much the user wants a suggestion to stray from a center point of relevant choices, the first setting being different from a second setting provided by the user for a second group from the set of groups;

providing, by the one or more processors, at least one first suggestion from the set of ranked suggestions to the user;

receiving implicit feedback from the user in response to providing the at least one first suggestion, the implicit feedback including positive feedback; and

responsive to receiving the positive feedback, providing at least one second suggestion from the set of ranked suggestions to the user that stray from the center point of relevant choices for the user.

2. The method of claim 1 further comprising:

wherein the implicit feedback is negative feedback from the user based on the at least one first suggestion; and

responsive to receiving the negative feedback, providing the at least one second suggestion for the user that is closer to the center point of relevant choices for the user.

3. The method of claim 1 wherein the query for content includes the user interest and a level of relevancy for the content as compared to the user interest.

4. The method of claim 1 wherein the set of ranked suggestions is also based on at least one of a recency of user interests, frequency of actions associated with the user interests and social affinity of the user to other users that generated any of the at least one suggestions.

5. The method of claim 1, wherein the implicit feedback is determined from at least one of time spent looking at the suggestions, a number of suggestions selected, relevance of the selected suggestions and number of approvals or disapprovals of the selected suggestions.

6. The method of claim 1, further comprising generating a user interface that includes a sliding scale for the user to specify the first setting.

7. The method of claim 1 wherein a probability of straying by a certain amount from the relevant choice is an inverse exponential function.

8. The method of claim 1, wherein the implicit feedback is normalized based on previous counts of implicit feedback received.

9. A system comprising:

one or more processors;

a user activity processing unit stored on a memory and executable by the one or more processors, the user activity processing unit configured to identify a plurality of user interests from user activity associated with a user, apply a weight to each of the user interests based on a frequency of user activity related to each of the user interests and to generate groups on a social network for the user, at least a first group from the set of the groups including a user interest;

a recommendation engine stored on the memory and executable by the one or more processors, the recommendation engine coupled to the user activity processing unit and configured to generate a query for content for the first group that includes the user interest and to receive the content for the first group;

a ranking engine stored on the memory and executable by the one or more processors, the ranking engine coupled to the recommendation engine and configured to generate a set of ranked suggestions from the content for the first group based on relevancy to the user interest, the weight of the user interest and explicit feedback from the user, the explicit feedback including a first setting for how much the user wants a suggestion to stray from a center point of relevant choices, the first setting being different from a second setting provided by the user for a second group from the set of groups; and

a simulated annealing engine stored on the memory and executable by the one or more processors, the simulated annealing engine coupled to the ranking engine and configured to provide at least one first suggestion from the set of ranked suggestions to the user, to receive implicit feedback from the user in response to providing the at least one first suggestion, the implicit feedback including positive feedback and responsive to receiving the positive feedback, to provide at least one second suggestions from the set of ranked suggestions for the user that stray from the center point of relevant choices for the user.

10. The system of claim 9 wherein the implicit feedback is negative feedback from the user based on the at least one first suggestion and responsive to receiving the negative feedback, the simulated annealing engine provides at least one second suggestion for the user that is closer to the center point of relevant choices for the user.

11. The system of claim 9 wherein the query for content includes the user interest and a level of relevancy for the content as compared to the user interest.

12. The system of claim 9 wherein the set of ranked suggestions is also based on at least one of a recency of user interests, frequency of actions associated with the user interests and social affinity of the user to other users that generated any of the at least one suggestions.

13. The system of claim 10 wherein the implicit feedback is determined from at least one of time spent looking at the suggestions, a number of suggestions selected, relevance of the selected suggestions and a number of approvals or disapprovals of the selected suggestions.

14. The system of claim 10 further comprising a graphical user interface (GUI) engine generating a user interface that includes a sliding scale for the user to specify the first setting.

15. A computer program product stored on a non-transitory computer readable medium including a computer readable program, wherein the computer readable program when executed on a computer causes the computer to:

identify a plurality of user interests from user activity associated with a user;

apply a weight to each of the user interests based on a frequency of user activity related to each of the user interests;

generate groups on a social network for the user, at least a first group from the set of the groups including a user interest;

generate a query for content for the first group that includes the user interest;

receive the content for the first group;

generate a set of ranked suggestions from the content for the first group based on relevancy to the user interest, the weight of the user interest and explicit feedback from the user, the explicit feedback including a first setting for how much the user wants a suggestion to stray from a center point of relevant choices, the first setting being different from a second setting provided by the user for a second group from the set of groups;

provide the at least one first suggestion from the set of ranked suggestions to the user;

receive implicit feedback from the user in response to providing the at least one first suggestion, the implicit feedback including positive feedback; and

responsive to receiving the positive feedback, provide at least one second suggestion from the set of ranked suggestions for the user that stray from the center point of relevant choices for the user.

16. The computer program product of claim 15, wherein the implicit feedback is negative feedback from the user based on the at least one first suggestion and responsive to receiving the negative feedback, providing the at least one second suggestion for the user that is closer to the center point of relevant choices for the user.

17. The computer program product of claim 15 wherein the query for content includes the user interest and a level for the content as compared to the user interest.