Systems and methods for fraud detection based on image analysis

1. A computer-implemented method comprising:

receiving, by a computing system, an advertisement including an image;

training, by the computing system, a machine learning model as a fraud classifier;

applying, by the computing system, the fraud classifier to the advertisement to generate a fraud assessment value for the advertisement;

training, by the computing system, a machine learning model as an image classifier;

applying, by the computing system, the image classifier to the image to generate an image assessment value for the image indicating a probability that the image reflects predetermined subject matter associated with fraudulent advertisements;

comparing, by the computing system, the fraud assessment value and a threshold value for fraud assessment;

comparing, by the computing system, the image assessment value and a threshold value for image assessment; and

determining, by the computing system, fraud associated with the advertisement based on the comparing the fraud assessment value and the comparing the image assessment value, wherein the determining fraud comprises:

determining whether the fraud assessment value satisfies the threshold value for fraud assessment; and

determining whether the image assessment value satisfies the threshold value for image assessment.

2. The computer-implemented method of claim 1 , wherein the advertisement is proposed for publication within a social networking system.

3. The computer-implemented method of claim 1 , wherein

the fraud assessment value indicates a probability that the advertisement is fraudulent.

4. The computer-implemented method of claim 1 , further comprising:

providing for presentation selectable tags corresponding to predetermined subject matter associated with fraudulent advertisements; and

receiving an association of at least one tag relating to the image to train the image classifier.

5. The computer-implemented method of claim 1 , wherein the predetermined subject matter includes luxury goods.

6. The computer-implemented method of claim 1 , wherein the image classifier is trained based on images and related tags associated with more than advertisements.

7. The computer-implemented method of claim 1 , wherein the image classifier is trained on the fly based on images and related tags associated with advertisements only.

8. The computer-implemented method of claim 7 , wherein the images and related tags are associated by manual review.

9. A system comprising:

at least one processor; and

a memory storing instructions that, when executed by the at least one processor, cause the system to perform:

receiving an advertisement including an image;

training a machine learning model as a fraud classifier;

applying the fraud classifier to the advertisement to generate a fraud assessment value for the advertisement;

training a machine learning model as an image classifier;

applying the image classifier to the image to generate an image assessment value for the image indicating a probability that the image reflects predetermined subject matter associated with fraudulent advertisements;

comparing the fraud assessment value and a threshold value for fraud assessment;

comparing the image assessment value and a threshold value for image assessment; and

determining fraud associated with the advertisement based on the comparing the fraud assessment value and the comparing the image assessment value, wherein the determining fraud comprises:

determining whether the fraud assessment value satisfies the threshold value for fraud assessment; and

determining whether the image assessment value satisfies the threshold value for image assessment.

10. The system of claim 9 , wherein the advertisement is proposed for publication within a social networking system.

11. The system of claim 9 ,

the fraud assessment value indicates a probability that the advertisement is fraudulent.

12. The system of claim 9 , wherein the predetermined subject matter includes luxury goods.

13. A non-transitory computer-readable storage medium including instructions that, when executed by at least one processor of a computing system, cause the computing system to perform a method comprising:

receiving an advertisement including an image;

training a machine learning model as a fraud classifier;

applying the fraud classifier to the advertisement to generate a fraud assessment value for the advertisement;

training a machine learning model as an image classifier;

applying the image classifier to the image to generate an image assessment value for the image indicating a probability that the image reflects predetermined subject matter associated with fraudulent advertisements;

comparing the fraud assessment value and a threshold value for fraud assessment;

comparing the image assessment value and a threshold value for image assessment; and

determining fraud associated with the advertisement based on the comparing the fraud assessment value and the comparing the image assessment value, wherein the determining fraud comprises:

determining whether the fraud assessment value satisfies the threshold value for fraud assessment, and

determining whether the image assessment value satisfies the threshold value for image assessment.

14. The non-transitory computer-readable storage medium of claim 13 , wherein the advertisement is proposed for publication within a social networking system.

15. The non-transitory computer-readable storage medium of claim 13 , wherein

the fraud assessment value indicates a probability that the advertisement is fraudulent.

16. The non-transitory computer-readable storage medium of claim 13 , wherein the predetermined subject matter includes luxury goods.