Predicting a concept prediction for media

1. A method comprising:

generating, by use of a processor, a review job comprising at least one review question and a corresponding expert concept prediction for an expert media set that is a subset of a media corpus;

generating media reviews from a plurality of reviewers using the review job for the expert media set;

extracting media features for media of the media reviews;

training a review model with the media reviews and corresponding media features;

producing the concept prediction for media of the media corpus using the review model; and

selecting media based on the predicted concept prediction for a media set.

2. The method of claim 1 , wherein generating the review job comprises:

presenting media of the expert media set to an expert;

capturing the expert concept prediction;

standardizing a preference format for the expert concept prediction; and

designing the review job based on a plurality of expert concept predictions.

3. The method of claim 2 , wherein generating the review job further comprises:

presenting the review job to a plurality of reviewers;

determining whether an agreement threshold is satisfied for concept predictions received from the reviewers; and

redesigning the review job in response to the agreement threshold not being satisfied.

4. The method of claim 1 , wherein generating the media reviews comprises:

presenting the review job to the plurality of reviewers;

receiving concept predictions for the review question of the review job from the plurality of reviewers;

determining whether the concept predictions satisfy an accuracy threshold; and

recording the media review in response to the accuracy threshold being satisfied, wherein each media review comprises a consensus concept prediction.

5. The method of claim 4 , wherein generating the media review further comprises:

regenerating the review job in response to the accuracy threshold not being satisfied for the plurality of reviewers; and

removing reviewers from the plurality of reviewers with concept predictions that do not satisfy the accuracy threshold.

6. The method of claim 4 , wherein generating the media reviews further comprises identifying a plurality of reviewers based answers to the review job for the expert media set from the plurality of reviewers.

7. The method of claim 1 , wherein generating the training the review model comprises:

extracting media features from media;

modifying model parameters for the review model; and

training a given review model based on the modified model parameters;

comparing the given review model to a current review model; and

selecting the given review model as the current review model in response to improvement.

8. The method of claim 1 , the method further comprising performing a constrained optimization calculation for presenting the media.

9. The method of claim 8 , wherein performing the constrained optimization calculation comprises:

defining budget requirements for presenting the media;

defining an objective function for presenting the media;

defining constraints comprising a placement number and a desired spend from the budget requirements; and

calculating the constrained optimization.

10. The method of claim 1 , the method further comprising presenting a promotion with the selected media.

11. The method of claim 1 , the method further comprising:

receiving a target medium;

predicting the concept prediction for the target medium; and

calculating a bid for the target medium based on the concept prediction.

12. The method of claim 1 , the method further:

selecting a media sourcing strategy from the group consisting of a random strategy, an active learning strategy, and a targeted strategy, wherein the random strategy selects media at random, the active learning strategy selects media based on the concept prediction and interactive queries directed to reviewers, and the target strategy selects media based on media metrics; and

retraining the review model with media selected using the media sourcing strategy.

13. The method of claim 1 , wherein extracting media features comprises:

receiving media metrics for media;

encoding a media category for the media;

encoding a text description for the media;

encoding an image for the media;

encoding text from audio of the media; and

encoding video of the media.

14. An apparatus comprising:

a processor;

a memory that stores code executable by the processor to perform:

generating a review job comprising at least one review question and a corresponding expert concept prediction for an expert media set that is a subset of a media corpus;

generating media reviews from a plurality of reviewers using the review job for the expert media set;

extracting media features for media of the media reviews;

training a review model with the media reviews and corresponding media features;

producing the concept prediction for media of the media corpus using the review model; and

selecting media based on the predicted concept prediction for a media set.

15. The apparatus of claim 14 , wherein generating the review job comprises:

presenting media of the expert media set to an expert;

capturing the expert concept prediction;

standardizing a preference format for the expert concept prediction; and

designing the review job based on a plurality of expert concept predictions.

16. The apparatus of claim 15 , wherein generating the review job further comprises:

presenting the review job to a plurality of reviewers;

determining whether an agreement threshold is satisfied for concept predictions received from the reviewers; and

redesigning the review job in response to the agreement threshold not being satisfied.

17. The apparatus of claim 14 , wherein generating the media reviews comprises:

presenting the review job to the plurality of reviewers;

receiving concept predictions for the review question of the review job from the plurality of reviewers;

determining whether the concept predictions satisfy an accuracy threshold; and

recording the media review in response to the accuracy threshold being satisfied, wherein each media review comprises a consensus concept prediction.

18. The apparatus of claim 17 , wherein generating the media review further comprises:

regenerating the review job in response to the accuracy threshold not being satisfied for the plurality of reviewers; and

removing reviewers from the plurality of reviewers with concept predictions that do not satisfy the accuracy threshold.

19. The apparatus of claim 17 , wherein generating the media reviews further comprises identifying a plurality of reviewers based answers to the review job for the expert media set from the plurality of reviewers.

20. A program product comprising a non-transitory storage medium storing code executable by the processor to perform:

generating a review job comprising at least one review question and a corresponding expert concept prediction for an expert media set that is a subset of a media corpus;

generating media reviews from a plurality of reviewers using the review job for the expert media set;

extracting media features for media of the media reviews;

training a review model with the media reviews and corresponding media features;

producing the concept prediction for media of the media corpus using the review model; and

selecting media based on the predicted concept prediction for a media set.