Adversarial method and system for generating user preferred contents

1. A recommendation method, comprising:

retrieving content consumption data comprising content consumed and content not consumed;

identifying, based on the content consumption data, a first piece of content not consumed;

determining a first feature of the first piece of content, wherein the first feature has a first feature value related to negative consumption of the first piece of content;

disguising the first piece of content as a second piece of content by changing, using a first system, the first feature of the first piece content from the first feature value to a second feature value different from the first feature value;

determining whether the disguising is successful by classifying, using a second system that is different from the first system, the second piece of content, wherein the disguising is successful in response to the second piece of content classified with a class representing positive consumption; and

providing, to an electronic device, the second piece of content in response to determining the disguising is successful;

wherein the first system is trained to disguise pieces of content that are not consumed such that the second system classifies the resulting disguised pieces of content with the class representing positive consumption, and the second system is trained to mimic user behavior on content consumption rating.

2. The recommendation method of claim 1 , further comprising:

transforming, using the first system, the first piece of content to a plurality of pieces of content, wherein each piece of the plurality of pieces of content changes a distinctive feature of the first piece of content;

evaluating, using the second system, each piece of the plurality of pieces of content using a trained model, wherein the trained model has been trained based on the content consumption data; and

identifying the second piece of content based on an evaluation result determining the second piece of content has a likelihood of consumption success.

3. The recommendation method of claim 2 , further comprising:

updating the first system based on the trained model.

4. The recommendation method of claim 2 , further comprising:

classifying, by the second system, the second piece of content with a positive consumption class representing positive consumption or a negative consumption class representing negative consumption; and

upon a determination that the second piece of content is classified with the positive consumption class, providing the second piece of content to the electronic device.

5. The recommendation method of claim 2 , wherein the first system comprises a first neural network, and the second system comprises a second neural network.

6. The recommendation method of claim 5 , further comprising:

optimizing, by an adversarial process, a likelihood of consumption of the second piece of content based on evaluating a disguise objective for the first neural network and a discriminator objective from for the second neural network.

7. The recommendation method of claim 1 , further comprising:

identifying, based on the content consumption data, a third piece of content consumed;

determining a second feature that causes consumption of the third piece of content; and

finding the second piece of content, wherein the second piece of content replaces the first feature with the second feature.

8. An electronic device comprising:

a memory storing instructions; and

at least one processor executing the instructions including a process configured to:

retrieve content consumption data comprising content consumed and content not consumed;

identify, based on the content consumption data, a first piece of content not consumed;

determine a first feature of the first piece of content, wherein the first feature has a first feature value related to negative consumption of the first piece of content;

disguise the first piece of content as a second piece of content by changing, using a first system, the first feature of the first piece content from the first feature value to a second feature value different from the first feature value;

determine whether the disguise is successful by classifying, using a second system that is different from the first system, the second piece of content, wherein the disguise is successful in response to the second piece of content classified with a class representing positive consumption; and

provide the second piece of content in response to determining the disguise is successful;

wherein the first system is trained to disguise pieces of content that are not consumed such that the second system classifies the resulting disguised pieces of content with the class representing positive consumption, and the second system is trained to mimic user behavior on content consumption rating.

9. The electronic device of claim 8 , wherein the process is further configured to:

transform, using the first system, the first piece of content to a plurality of pieces of content, wherein each piece of the plurality of pieces of content changes a distinctive feature of the first piece of content;

evaluate, using the second system, each piece of the plurality of pieces of content using a trained model, wherein the trained model has been trained based on the content consumption data; and

identify the second piece of content based on an evaluation result determining the second piece of content has a likelihood of consumption success.

10. The electronic device of claim 9 , wherein the process is further configured to:

update the first system based on the trained model;

classify, by the second system, the second piece of content with a positive consumption class representing positive consumption or a negative consumption class representing negative consumption; and

upon a determination that the second piece of content is classified with the positive consumption class, provide the second piece of content to the electronic device.

11. The electronic device of claim 9 , wherein the first system comprises a first neural network, and the second system comprises a second neural network.

12. The electronic device of claim 11 , wherein the process is further configured to:

optimize, by an adversarial process, a likelihood of consumption of the second piece of content based on evaluating a disguise objective for the first neural network and a discriminator objective for the second neural network.

13. The electronic device of claim 8 , wherein the process is further configured to:

identify, based on the content consumption data, a third piece of content consumed;

determine a second feature that causes consumption of the third piece of content; and

find the second piece of content, wherein the second piece of content replaces the first feature with the second feature.

14. A non-transitory processor-readable medium that includes a program that when executed by a processor performing a method comprising:

retrieving content consumption data comprising content consumed and content not consumed;

identifying, based on the content consumption data, a first piece of content not consumed;

determining a first feature of the first piece of content, wherein the first feature has a first feature value related to negative consumption of the first piece of content;

disguising the first piece of content as a second piece of content by changing, using a first system, the first feature of the first piece content from the first feature value to a second feature value different from the first feature value;

determining whether the disguising is successful by classifying, using a second system that is different from the first system, the second piece of content, wherein the disguising is successful in response to the second piece of content classified with a class representing positive consumption; and

providing, to an electronic device, the second piece of content in response to determining the disguising is successful;

wherein the first system is trained to disguise pieces of content that are not consumed such that the second system classifies the resulting disguised pieces of content with the class representing positive consumption, and the second system is trained to mimic user behavior on content consumption rating.

15. The non-transitory processor-readable medium of claim 14 , wherein the method further comprises:

transforming, using the first system, the first piece of content to a plurality of pieces of content, wherein each piece of the plurality of pieces of content changes a distinctive feature of the first piece of content;

evaluating, using the second system, each piece of the plurality of pieces of content using a trained model, wherein the trained model has been trained based on the content consumption data; and

identifying the second piece of content based on an evaluation result determining the second piece of content has a likelihood of consumption success.

16. The non-transitory processor-readable medium of claim 15 , wherein the method further comprises:

updating the first system based on the trained model;

classifying, by the second system, the second piece of content with a positive consumption class representing positive consumption or a negative consumption class representing negative consumption; and

upon a determination that the second piece of content is classified with the positive consumption class, providing the second piece of content to the electronic device.

17. The non-transitory processor-readable medium of claim 15 , wherein the first system comprises a first neural network, and the second system comprises a second neural network.

18. The non-transitory processor-readable medium of claim 17 , wherein the method is further configured to:

optimize, by an adversarial process, a likelihood of consumption of the second piece of content based on evaluating a disguise objective for the first neural network and a discriminator objective for the second neural network.

19. The non-transitory processor-readable medium of claim 14 , wherein the method further comprises:

identifying, based on the content consumption data, a third piece of content consumed;

determining a second feature that causes consumption of the third piece of content; and

finding the second piece of content, wherein the second piece of content replaces the first feature with the second feature.