Identifying a thumbnail image to represent a video

A system comprising:

• a memory;
• and a processor, operatively coupled to the memory, the processor to:
• identify a plurality of candidate thumbnails for a media item;
• determine, for each candidate thumbnail, characteristics comprising two or more of a click through rate, a watch completion number or a generated revenue amount based on actions of a plurality of users over a period of time in relation to a respective candidate thumbnail during display of the respective candidate thumbnail on a plurality of user devices of the plurality of users;
• determine, using the characteristics of each candidate thumbnail, a success ranking of each candidate thumbnail in view of target metrics, wherein the target metrics relate to consumption of the media item in response to the candidate thumbnail and comprise two or more of a target click through rate, a target watch completion number or a target generated revenue amount;
• use a probabilistic selection learning utility to probabilistically select a set of candidate thumbnails from the plurality of candidate thumbnails,
• wherein each candidate thumbnail in the set is selected with a probability that is determined based on a success ranking of a respective candidate thumbnail from the set and success rankings of other candidate thumbnails of the plurality of candidate thumbnails,
• wherein the probabilistic selection learning utility is continuously trained with the set of candidate thumbnails selected from the plurality of candidate thumbnails using a metaheuristic learning process comprising a first function and a second function each associated with a different prioritization of success rankings;
• provide for display each selected candidate thumbnail from the set of candidate thumbnails as determined by the trained probabilistic selection learning utility;
• update the probabilistic selection learning utility with a success ranking of each displayed candidate thumbnail from the set in view of consumption of the media item in response to the displayed candidate thumbnail;
• select an optimum thumbnail for the media item based on updated success rankings of the displayed candidate thumbnails as determined by the trained probabilistic selection learning utility.

A non-transitory computer-readable medium storing executable instructions that, when executed by a computing, device, cause the computing device to:

• identify a plurality of candidate thumbnails for a media item;
• determine, for each candidate thumbnail, characteristics comprising two or more of a click through rate, a watch completion number or a generated revenue amount based on actions of a plurality of users over a period of time in relation to a respective candidate thumbnail during display of the respective candidate thumbnail on a plurality of user devices of the plurality of users;
• determine, using the characteristics of each candidate thumbnail, a success ranking of each candidate thumbnail in view of target metrics wherein the target metrics relate to consumption of the media item in response to the candidate thumbnail and comprise two or more of a target click through rate, a target watch completion number or a target generated revenue amount;
• use, by the computing device, a probabilistic selection learning utility to probabilistically select a set of candidate thumbnails from the plurality of candidate thumbnails,
• wherein each candidate thumbnail in the set is selected with a probability that is determined based on a success ranking of a respective candidate thumbnail from the set and success rankings of other candidate thumbnails of the plurality of candidate thumbnails,
• wherein the probabilistic selection learning utility is continuously trained with the set of candidate thumbnails selected from the plurality of candidate thumbnails using a metaheuristic learning process comprising a first function and a second function each associated with a different prioritization of success rankings;
• provide for display each selected candidate thumbnail from the set of candidate thumbnails as determined by the trained probabilistic selection learning utility;
• update the probabilistic selection learning utility with a success ranking of each displayed candidate thumbnail from the set in view of consumption of the media item in response to the displayed candidate thumbnail;
• and
• select an optimum thumbnail for the media, item based on updated success rankings of the displayed candidate thumbnails as determined by the trained probabilistic selection learning utility.