Media content enhancement based on user feedback of multiple variations
Various disclosed embodiments are directed to using one or more algorithms or models to select a suitable or optimal variation, among multiple variations, of a given content item based on feedback. Such feedback guides the algorithm or model to arrive at suitable variation result such that the variation result is produced as the output for consumption by users. Further, various embodiments resolve tedious manual user input requirements and reduce computing resource consumption, among other things, as described in more detail below.
1. A non-transitory computer readable medium storing computer-usable instructions that, when used by one or more processors, cause the one or more processors to perform operations comprising:
receiving a media content item that is an image of a plurality of pixel values; converting the media content item into a first feature vector representative of the plurality of pixel values in search space:
receiving an indication that a user has set a boundary or range of parameter values for which a model will generate variations from:
based on the boundary or range and the first feature vector, automatically generating a plurality of variations of the media content item by automatically changing in the search space, the first feature vector representative of a change in at least one of, vibrance, saturation, brightness, contrast, or sharpness of one or more of the plurality of pixel values, each variation, of the plurality of variations, being a different version of the image; receiving explicit user feedback for each variation of the plurality of variations, wherein the explicit user feedback corresponds to a scaled user rating of a respective variation of the plurality of variations according to an aesthetic preference for the respective variation; based on the explicit user feedback, scoring each variation of the plurality of variations according to the scaled user rating of the respective variation; based on the changing, in the search space, the first feature vector and the scoring of each variation, automatically generating, via a Bayesian Optimization Model a first variation of the image based on using a surrogate function that models an objective function representing a true distribution of user feedback for the plurality of variations by sampling, in the search space, at least a second feature vector representing the first variation based on minimizing a distance to the objective function evaluated at a maximum, wherein the first variation represents the maximum of the objective function and the maximum indicates a highest scoring variation according to the user feedback; based on the generating of the first variation, generating an output image of pixel values that represent the first variation; and based on the generating of the output image, causing
presentation, at a computing device associated with the user, of the output image.
2. The non-transitory computer readable medium of claim 1 , wherein the user feedback includes a user rating of a particular variation, of the plurality of variations, based on showing the particular variation to a second user and prompting the second user to rate the particular variation.
3. The non-transitory computer readable medium of claim 1 , wherein the generating of the first variation is based further on receiving additional user feedback includes user input at a user interface, and wherein the additional user feedback includes at least one of: a purchase of an item associated with the variation, a click of a button at a user interface page, and a view of the user interface page.
4. The non-transitory computer readable medium of claim 1 , wherein the media content item includes a video, and wherein the first variation includes a Graphics Interchange Format (GIF) of the video.
5. The non-transitory computer readable medium of claim 1 , wherein the one or more processors are caused to perform further operations comprising:
subsequent to the generating of the first variation, generating, via the model, a second set of variations that resemble the media content item; based on the generating of the second set of variations, receiving second user feedback for each variation of the second set of variations; based on the second user feedback, scoring each variation of the second set of variations; based on the scoring of the second set of variations, generating a second variation; and selecting the first variation instead of the second variation based on a score of the first variation being higher relative to the second variation, and wherein the causing presentation of the output image is further based on the selecting of the first variation.
6. The non-transitory computer readable medium of claim 1 , wherein the causing presentation includes causing the the output image to be produced at an application page as part of a digital marketing campaign.
7. The non-transitory computer readable medium of claim 1 , wherein each of the plurality of variations include a set pixels with different values relative to a corresponding set of pixels of the image.
8. The non-transitory computer readable medium of claim 1 , wherein each variation of plurality of variations include a set of pixels representing real world objects that are not included in the media content item.
9. The non-transitory computer readable medium of claim 1 , wherein the generating of the first variation is part of training or optimizing the model, and wherein the one or more processors further include operations comprising:
receiving, subsequent to the generating of the first variation, a first media content item; determining the plurality of variations; and changing first one or more image data-values of the first media content item to second one or more image data-values based on the scoring of each variation of the plurality of variations.
10. A computerized system, comprising:
at least one processor; and
at least one computer readable storage medium storing computer instructions that when executed by the at least one processor cause the at least one processor to perform operations comprising:
receiving a first media content item, the first media content item being an image or set of images that include a plurality of pixel values;
converting the first media content item into a first feature vector representative of the plurality of pixel values in search space:
generating a plurality of variations associated with the first media content item or a second media content item based at least in part on the converting and changing, in the search space, one or more values of the first feature vector, each variation, of the plurality of variations, resembling the image or set of images except each variation includes a change in at least one of, vibrance, saturation, brightness, contrast, or sharpness of one or more of the plurality of pixel values;
determining a score for each variation of the plurality of variations based on explicit user feedback for each variation, wherein the explicit user feedback corresponds to a scaled user rating of aesthetic preference for a respective variation of the plurality of variations according to one or more users;
changing first one or more pixel values of the plurality of pixel values to second one or more pixel values based on the explicit user feedback and using a surrogate function that models an objective function representing a true distribution of user feedback for the plurality of variations by sampling, in the search space, at least a second feature vector representing a first variation based on minimizing a distance to the objective function evaluated at a maximum;
based at least in part on the changing of the first one or more pixel values to the second one or more pixel values, generating an output image of pixel values that represents the first variation; and
causing presentation, at a computing device associated with a user, of the output image.
11. A computer-implemented method comprising:
receiving a media content item that is an image of a plurality of pixel values;
converting the media content item into a first feature vector representative of the plurality of pixel values in search space;
receiving an indication that a user has set a boundary or range of parameter values for which a model will generate variations from
based at least in part on the first feature vector and the boundary or range, generating a plurality of variations of media content item by changing one or more values of the first feature vector in the search space, each variation, of the plurality of variations, resembles the media content item except that each variation has one or more of, vibrance, saturation, brightness, contrast, or sharpness that are not included in the image or set of images;
receiving explicit user feedback for each variation, of the plurality of variations, wherein the explicit user feedback corresponds to a scaled user rating of aesthetic preference for a respective variation of the plurality of variations according to one or more users;
based on the explicit user feedback, scoring each variation of the plurality of variations according to the scaled user rating of the respective variation;
based on the scoring and the changing of the one or more values of the first feature vector, sampling, via the model and in the search space, a second feature vector representing a first variation of the image or set of images based at least in part on using a surrogate function that models an objective function representing a true distribution of user feedback for the plurality of variations and further based on minimizing a distance to the objective function evaluated at a maximum;
and based on the sampling of the first feature vector generating an output image of pixel values that represent the first variation.
12. The computer-implemented method of claim 11 , wherein the model is a Bayesian Optimization algorithm, that approximates a distribution of the objective function With the plurality of variations, and wherein the maximum indicating a highest scoring variant according to the explicit user feedback.
13. The computer-implemented method of claim 11 , wherein the explicit user feedback includes a user rating of a particular variation, of the plurality of variations, based on showing the particular variation to a second user and prompting the second user to rate the particular variation.
14. The computer-implemented method of claim 11 , wherein the media content item includes a video, and wherein the first variation includes a Graphics Interchange Format (GIF) of the video.
15. The computer-implemented method of claim 11 , further comprising:
in response to the generating of the first variation, determining a second set of variations of the plurality of variations, the determining of the second set of variations being part of a function that approximates a distribution associated with the user feedback;
based on the determining of the second set of variations, receiving second user feedback for each variation of the second set of variations;
based on at least a portion of the second user feedback, scoring each variation of the second set of variations;
based on the scoring of the second set of variations, determining a second variation; and
selecting the first variation instead of the second variation based on a score of the first variation being higher, and wherein the causing presentation of the first variation is further based on the selecting of the first variation.
16. The non-transitory computer readable medium of claim 10 , wherein the surrogate function that models the objective function representing a true distribution is based on using the Bayesian Optimization model that models the objective function associated with the user feedback includes generating the first variation using the model that approximates a ground truth distribution of user feedback scores, wherein the user feedback represents only a portion of the ground truth distribution of user feedback scores.
17. The computerized system of claim 10 , wherein the changing is based on using a model that has been trained or optimized, prior to the receiving of the first media content item, based on the user feedback, and wherein the plurality of variations are variations of the second media content item.