STATISTICAL FEATURE ENGINEERING OF USER ATTRIBUTES

• US 20170262445A1
• Filed: 03/08/2016
• Published: 09/14/2017
• Est. Priority Date: 03/08/2016
• Status: Active Grant

• Current Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Original Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Inventors: Jeon, Sung-eok, Husain, Aliasgar
• Granted Patent: US 10,509,791 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 16/24578 using ranking
• CPC Class Codes: G06F 16/337 Profile generation, learning or modification
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: H04L 67/306 User profiles
• CPC Class Codes: H04L 67/53 using third party service providers
• CPC Class Codes: H04L 67/535 Tracking the activity of the user network monitoring arrangements H04L43/00; recording of computer activity G06F11/34

Abstract

A method trains a model for providing content items to users of a social networking system. The system generates profile vectors based on user profile information such as demographic data and personal data. The system logs actions performed by users on the social networking system and generates behavior vectors based on the logged actions. The profile vectors and behavior vectors are each associated with a user attribute, e.g., the age or gender of a user. The system generates a difference vector based on a profile vector and a behavior vector. The difference vector is then used as a feature to train the model using machine learning techniques. The trained model may select content items that a target user is most likely to be interested in and interact with.

Claims

1. A method comprising:

receiving user profile information for a user of a social networking system;

generating a profile vector associated with a user attribute based on the user profile information, the profile vector comprising profile bin scores for each of two or more attribute bins in a distribution of values of the user attribute, each attribute bin corresponding to a range of values in the distribution of values and each profile bin score indicating how closely the user profile information associated with the user attribute matches attributes characteristic of the ranges corresponding to the two or more attribute bins;

maintaining a plurality of actions performed by the user in the social networking system;

generating a behavior vector associated with the user attribute based on the plurality of actions performed by the user, the behavior vector comprising behavior bin scores for the each of the two or more attribute ranges corresponding to the two or more attribute bins for the user attribute, each behavior bin score indicating how closely the plurality of actions performed by the user associated with the user attribute matches attributes characteristic of the ranges corresponding to the two or more attribute bins;

generating a difference vector based on the difference between the profile vector and the behavior vector, the difference vector comprising difference scores for the each of the two or more attribute ranges corresponding to the two or more attribute bins; and

training a model configured for providing content to the user of the social networking system based at least in part on the user attribute, the model trained using the difference vector as a feature.

2. The method of claim 1, wherein the profile vector has two or more dimensions, each dimension of the two or more dimensions associated with a different user attribute, wherein the behavior vector has the two or more dimensions, and wherein the difference vector has the two or more dimensions.

3. The method of claim 1, further comprising retraining the model periodically over time using one or more difference vectors.

4. The method of claim 1, wherein the profile bin scores are percentage values that sum to 100% and wherein the behavior bin scores are percentage values that sum to 100%.

5. The method of claim 1, further comprising:

generating a second difference vector including target population information describing users of the social networking system; and

providing content to the user based at least in part on the trained model and the second difference vector.

6. The method of claim 1, further comprising:

receiving a plurality of content items, each content item of the plurality of content items associated with a bid value;

determining a probability for each content item of the plurality of content items based at least in part on the trained model, the probability indicating a likelihood that the user will interact with the each content item of the plurality of content items;

determining an expected value for each content item of the plurality of content items based at least in part on the bid value corresponding to the each content item and the probability corresponding to the each content item; and

providing one or more content items of the plurality of content items to the user, the expected value of each content item of the one or more content items being greater than a threshold expected value.

7. The method of claim 6, wherein the content items include sponsored content, and wherein the bid values are represented by monetary amounts.

8. A computer program product comprising a computer readable storage medium having instructions encoded therein that, when executed by a processor, cause the processor to:

receive user profile information for a user of a social networking system;

generate a profile vector associated with a user attribute based on the user profile information, the profile vector comprising profile bin scores for each of two or more attribute bins in a distribution of values of the user attribute, each attribute bin corresponding to a range of values in the distribution of values and each profile bin score indicating how closely the user profile information associated with the user attribute matches attributes characteristic of the ranges corresponding to the two or more attribute bins;

maintain a plurality of actions performed by the user in the social networking system;

generate a behavior vector associated with the user attribute based on the plurality of actions performed by the user, the behavior vector comprising behavior bin scores for the each of the two or more attribute ranges corresponding to the two or more attribute bins for the user attribute, each behavior bin score indicating how closely the plurality of actions performed by the user associated with the user attribute matches attributes characteristic of the ranges corresponding to the two or more attribute bins;

generate a difference vector based on the difference between the profile vector and the behavior vector, the difference vector comprising difference scores for the each of the two or more attribute ranges corresponding to the two or more attribute bins; and

train a model configured for providing content to the user of the social networking system based at least in part on the user attribute, the model trained using the difference vector as a feature.

9. The computer program product of claim 8, wherein the profile vector has two or more dimensions, each dimension of the two or more dimensions associated with a different user attribute, wherein the behavior vector has the two or more dimensions, and wherein the difference vector has the two or more dimensions.

10. The computer program product of claim 8, further comprising instructions encoded therein that, when executed by the processor, cause the processor to retrain the model periodically over time using one or more difference vectors.

11. The computer program product of claim 8, wherein the profile bin scores are percentage values that sum to 100% and wherein the behavior bin scores are percentage values that sum to 100%.

12. The computer program product of claim 8, further comprising instructions encoded therein that, when executed by the processor, cause the processor to:

generate a second difference vector including target population information describing users of the social networking system; and

provide content to the user based at least in part on the trained model and the second difference vector.

13. The computer program product of claim 8, further comprising instructions encoded therein that, when executed by the processor, cause the processor to:

receive a plurality of content items, each content item of the plurality of content items associated with a bid value;

determine a probability for each content item of the plurality of content items based at least in part on the trained model, the probability indicating a likelihood that the user will interact with the each content item of the plurality of content items;

determine an expected value for each content item of the plurality of content items based at least in part on the bid value corresponding to the each content item and the probability corresponding to the each content item; and

provide one or more content items of the plurality of content items to the user, the expected value of each content item of the one or more content items being greater than a threshold expected value.

14. The computer program product of claim 13, wherein the content items include sponsored content, and wherein the bid values are represented by monetary amounts.

15. A system comprising a web server configured for:

receiving user profile information for a user of a social networking system;

generating a profile vector associated with a user attribute based on the user profile information, the profile vector comprising profile bin scores for each of two or more attribute bins in a distribution of values of the user attribute, each attribute bin corresponding to a range of values in the distribution of values and each profile bin score indicating how closely the user profile information associated with the user attribute matches attributes characteristic of the ranges corresponding to the two or more attribute bins;

maintaining a plurality of actions performed by the user in the social networking system;

generating a behavior vector associated with the user attribute based on the plurality of actions performed by the user, the behavior vector comprising behavior bin scores for the each of the two or more attribute ranges corresponding to the two or more attribute bins for the user attribute, each behavior bin score indicating how closely the plurality of actions performed by the user associated with the user attribute matches attributes characteristic of the ranges corresponding to the two or more attribute bins;

generating a difference vector based on the difference between the profile vector and the behavior vector, the difference vector comprising difference scores for the each of the two or more attribute ranges corresponding to the two or more attribute bins; and

training a model configured for providing content to the user of the social networking system based at least in part on the user attribute, the model trained using the difference vector as a feature.

16. The system of claim 15, wherein the profile vector has two or more dimensions, each dimension of the two or more dimensions associated with a different user attribute, wherein the behavior vector has the two or more dimensions, and wherein the difference vector has the two or more dimensions.

17. The system of claim 15, wherein the web server is further configured for retraining the model periodically over time using one or more difference vectors.

18. The system of claim 15, wherein the profile bin scores are percentage values that sum to 100% and wherein the behavior bin scores are percentage values that sum to 100%.

19. The system of claim 15, wherein the web server is further configured for:

generating a second difference vector including target population information describing users of the social networking system; and

providing content to the user based at least in part on the trained model and the second difference vector.

20. The system of claim 15, wherein the web server is further configured for:

receiving a plurality of content items, each content item of the plurality of content items associated with a bid value;

determining a probability for each content item of the plurality of content items based at least in part on the trained model, the probability indicating a likelihood that the user will interact with the each content item of the plurality of content items;

determining an expected value for each content item of the plurality of content items based at least in part on the bid value corresponding to the each content item and the probability corresponding to the each content item; and

providing one or more content items of the plurality of content items to the user, the expected value of each content item of the one or more content items being greater than a threshold expected value.

21. The system of claim 20, wherein the content items include sponsored content, and wherein the bid values are represented by monetary amounts.