Predicting users' attributes based on users' behaviors

• US 8,756,184 B2
• Filed: 12/01/2010
• Issued: 06/17/2014
• Est. Priority Date: 12/01/2009
• Status: Active Grant

• Current Assignee: Hulu LLC (The Walt Disney Company)
• Original Assignee: Hulu LLC (The Walt Disney Company)
• Inventors: Li, Hang, Zheng, Hua, Yang, Ting-hao, Tom, Richard W., Liao, Yupeng, Zhang, Qi
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): COUGHLAN, PETER D
• Application Number:
  US12/957,649
• Publication Number:
  US 20110131166A1
• Time in Patent Office: 1,294 Days
• Field of Search: 725/24
• US Class Current: 706/25
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/043 based on fuzzy logic, fuzzy membership or fuzzy inference, e.g. adaptive neuro-fuzzy inference systems [ANFIS]
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

A method, apparatus, system, article of manufacture, and computer readable storage medium provide the ability to predict and utilize a user'"'"'s attributes. A sample user behavior and a sample user attribute are collected. A model is trained based on the sample user behavior and sample user attribute. Using the model, a probability of a predicted user attribute based on the sample user behavior is predicted. Using the model and the probability, the predicted user attribute is fuzzily determined based on a real user behavior. The predicted user attribute is used to improve a user'"'"'s experience.

Claims

1. A computer-implemented method comprising:

collecting a plurality of sample user behaviors and a plurality of sample user attributes from a service that offers videos for viewing, wherein a sample user behavior is based on a sample user using the service and a sample user attribute is received from the sample user;

training a model to be able to produce the plurality of sample user attributes from the plurality of sample user behaviors;

after training, inputting the plurality of sample user behaviors into the trained model to produce predicted sample user attributes;

comparing the plurality of sample user attributes and the plurality of sample user behaviors to the predicted sample user attributes to determine associated probabilities for the predicted sample user attributes;

determining a real user behavior for a second user based on the second user using the service;

predicting, using the model, that the second user has a predicted user attribute of an associated probability based on the user having the real user behavior, wherein the predicted user attribute of the associated probability is not known for the second user; and

utilizing that the second user has the predicted user attribute of the associated probability to improve the second user'"'"'s experience using the service.

2. The computer-implemented method of claim 1, wherein the model comprises a neural network model.

3. The computer-implemented method of claim 2, wherein training the model comprises:

representing the sample user behavior as an input node having an input weight;

representing the sample user attribute as an output node having an output weight; and

determining one or more functions based using one or more processing nodes, wherein the one or more functions are configured to produce the output weight based on the input weight.

4. The computer-implemented method of claim 3, wherein training further comprises combining additional sample user behaviors and additional sample user attributes from multiple users with the sample user behavior and sample user attribute.

5. The computer-implemented method of claim 1, wherein comparing comprises:

comparing a distribution of the sample user attribute for one or more sample users to the predicted attribute to determine the associated probability for the predicted attribute based on a number of sample users that actually have the predicted attribute.

6. The computer-implemented method of claim 1, wherein comparing determines the associated probability based on a least squared fit analysis based on the predicted user attribute and sample user behavior.

7. The computer-implemented method of claim 1, wherein predicting comprises:

processing the real user behavior in the model to produce the predicted user attribute and a cumulative probability as the associated probability for the predicted user attribute.

8. The computer-implemented method of claim 1, wherein utilizing comprises transmitting a personalized recommendation to a user based on predicting that the second user has the predicted user attribute of the associated probability.

9. The computer-implemented method of claim 1, wherein the utilizing comprises transmitting a targeted advertisement to a user based on predicting that the second user has the predicted user attribute of the associated probability.

10. The computer-implemented method of claim 1, further comprising collecting a user'"'"'s feedback to improve the model by repeating the training using the user'"'"'s feedback.

11. The computer-implemented method of claim 1, wherein utilizing comprises selectively providing an advertisement to the second user based on the predicted user attribute of the associated probability.

12. The computer-implemented method of claim 11, wherein utilizing comprises charging an advertiser of an advertisement delivered to the second user while the second user is watching a video using the service differently based on the predicted user attribute of the associated probability.

13. An apparatus for utilizing a user'"'"'s predicted attributes in a computer system comprising:

(a) a computer having a memory;

(b) an application executing on the computer, wherein the application is configured to;

collect a plurality of sample user behaviors and a plurality of sample user attributes from a service that offers videos for viewing, wherein a sample user behavior is based on a sample user using the service and a sample user attribute is received from the sample user;

train a model to be able to produce the plurality of sample user attributes from the plurality of sample user behaviors;

after training, input the plurality of sample user behaviors into the trained model to produce predicted sample user attributes;

compare the plurality of sample user attributes and the plurality of sample user behaviors to the predicted sample user attributes to determine associated probabilities for the predicted sample user attributes;

determine a real user behavior for a second user based on the second user using the service;

predict, using the model, that the second user has a predicted user attribute of an associated probability based on the user having the real user behavior, wherein the predicted user attribute of the associated probability is not known for the second user; and

utilize that the second user has the predicted user attribute of the associated probability to improve the second user'"'"'s experience using the service.

14. The apparatus of claim 13, wherein the model comprises a neural network model.

15. The apparatus method of claim 14, wherein the application is configured to train a model by:

representing the sample user behavior as an input node having an input weight;

representing the sample user attribute as an output node having an output weight; and

determining one or more functions based using one or more processing nodes, wherein the one or more functions are configured to produce the output weight based on the input weight.

16. The apparatus of claim 15, wherein the application is further configured to train by combining additional sample user behaviors and additional sample user attributes from multiple users with the sample user behavior and sample user attribute.

17. The apparatus of claim 13, wherein the application is configured to compare by:

compare a distribution of the sample user attribute for one or more sample users to the predicted attribute to determine the associated probability for the predicted attribute based on a number of sample users that actually have the predicted attribute.

18. The apparatus of claim 13, wherein the application is configured to compare by determining the associated probability based on a least squared fit analysis based on the predicted user attribute and sample user behavior.

19. The apparatus of claim 13, wherein the application is configured to predict by:

processing the real user behavior in the model to produce the predicted user attribute and a cumulative probability as the associated probability for the predicted user attribute.

20. The apparatus of claim 13, wherein the application is configured to utilize by transmitting a personalized recommendation to a user based on predicting that the second user has the predicted user attribute of the associated probability.

21. The apparatus of claim 13, wherein the application is configured to utilize by transmitting targeted advertisement to a user based on predicting that the second user has the predicted user attribute of the associated probability.

22. The apparatus of claim 13, wherein the application is configured to further collect a user'"'"'s feedback to improve the model by repeating the training using the user'"'"'s feedback.

24. The non-transitory computer readable storage medium of claim 21, wherein the model comprises a neural network model.

25. The non-transitory computer readable storage medium of claim 24, wherein the training a model comprises:

representing the sample user behavior as an input node having an input weight;

representing the sample user attribute as an output node having an output weight; and

determining one or more functions based using one or more processing nodes, wherein the one or more functions are configured to produce the output weight based on the input weight.

26. The non-transitory computer readable storage medium of claim 25, wherein the training further comprises combining additional sample user behaviors and additional sample user attributes from multiple users with the sample user behavior and sample user attribute.

23. A non-transitory computer readable storage medium encoded with computer program instructions which when accessed by a computer cause the computer to load the program instructions to a memory therein creating a special purpose data structure causing the computer to operate as a specially programmed computer, executing a method of utilizing a user'"'"'s predicted attributes, comprising:

collecting, in the specially programmed computer, a plurality of sample user behaviors and a plurality of sample user attributes from a service that offers videos for viewing, wherein a sample user behavior is based on a sample user using the service and a sample user attribute is received from the sample user;

training, in the specially programmed computer, a model to be able to produce the plurality of sample user attributes from the plurality of sample user behaviors;

after training, inputting, in the specially programmed computer, the plurality of sample user behaviors into the trained model to produce predicted sample user attributes;

comparing, in the specially programmed computer, the plurality of sample user attributes and the plurality of sample user behaviors to the predicted sample user attributes to determine associated probabilities for the predicted sample user attributes;

determining, in the specially programmed computer, a real user behavior for a second user based on the second user using the service;

predicting, in the specially programmed computer, in the specially programmed computer, using the model, that the second user has a predicted user attribute of an associated probability based on the user having the real user behavior, wherein the predicted user attribute of the associated probability is not known for the second user; and

utilizing, in the specially programmed computer, that the second user has the predicted user attribute of the associated probability to improve the second user'"'"'s experience using the service.

27. The computer readable storage medium of claim 23, wherein the comparing comprises:

comparing a distribution of the sample user attribute for one or more sample users to the predicted attribute to determine the associated probability for the predicted attribute based on a number of sample users that actually have the predicted attribute.

28. The non-transitory computer readable storage medium of claim 23, wherein the comparing determines the associated probability based on a least squared fit analysis based on the predicted user attribute and sample user behavior.

29. The non-transitory computer readable storage medium of claim 23, wherein predicting comprises:

processing the real user behavior in the model to produce the predicted user attribute and a cumulative probability as the associated probability for the predicted user attribute.

30. The non-transitory computer readable storage medium of claim 23, wherein the utilizing comprises transmitting a personalized recommendation to a user based on predicting that the second user has the predicted user attribute of the associated probability.

31. The non-transitory computer readable storage medium of claim 23, wherein the utilizing comprises transmitting targeted advertisement to a user based on predicting that the second user has the predicted user attribute of the associated probability.

32. The non-transitory computer readable storage medium of claim 23, further comprising collecting a user'"'"'s feedback to improve the model by repeating the training using the user'"'"'s feedback.