Generating data to predict user affinity for a component to be included in a content item optimized for an online system user

A computer program product comprising a computer readable storage medium having instructions encoded thereon that, when executed by a processor, cause the processor to:

• receive, at an online system, a plurality of candidate content item components for including in one or more content items to be presented to a plurality of viewing users of the online system;
• perform a scoring process for a set of the plurality of candidate content item components, the scoring process comprising:
• generate a training content item comprising the set of the plurality of candidate content item components;
• present the training content item to one or more viewing users of the online system;
• receive information describing a performance metric associated with each impression of the content item, the information comprising one or more attributes associated with each of the one or more viewing users to whom the training content item was presented;
• and
• compute an affinity score of each of the one or more viewing users for each of the set of the plurality of candidate content item components based at least in part on the information describing the performance metric associated with each impression of the content item, the affinity score associated with the one or more attributes associated with each of the one or more viewing users to whom the training content item was presented;
• repeat the scoring process for a different set of the plurality of candidate content item components based at least in part on a heuristic, the heuristic based at least in part on a distribution of a set of affinity scores computed for each of the plurality of candidate content item components;
• identify an opportunity to present content to a subject user of the online system, the subject user associated with one or more additional attributes;
• identify one or more of the plurality of candidate content item components eligible to be included in an optimal content item for presentation to the subject user;
• and
• predict, using a machine-learned model, an affinity score of the subject user for each of the one or more candidate content item components based at least in part on the affinity scores for the set of the plurality of candidate content item components, the one or more attributes associated with the viewing users, and the one or more additional attributes associated with the subject user.

A method comprising:

• for each of a plurality of candidate content item components eligible to be included in one or more content items to be presented to a plurality of viewing users of an online system, performing a scoring process, the scoring process comprising:
• generating a training content item comprising a candidate content item component of the plurality of candidate content item components;
• presenting the training content item to one or more viewing users of the online system;
• receiving information describing a performance metric associated with each impression of the content item, the information comprising one or more attributes associated with each of the one or more viewing users to whom the training content item was presented;
• and
• computing an affinity score of each of the one or more viewing users for each of the plurality of candidate content item components based at least in part on the information describing the performance metric associated with each impression of the content item, the affinity score associated with the one or more attributes associated with each of the one or more viewing users to whom the training content item was presented;
• repeating the scoring process for one or more of the plurality of candidate content item components based at least in part on a distribution of a set of affinity scores computed for each of the plurality of candidate content item components;
• identifying an opportunity to present content to a subject user of the online system, the subject user associated with one or more additional attributes;
• identifying one or more of the plurality of candidate content item components eligible to be included in an optimal content item for presentation to the subject user;
• and
• predicting, using a machine-learned model, an affinity score of the subject user for each of the one or more candidate content item components based at least in part on the affinity scores for the set of the plurality of candidate content item components, the one or more attributes associated with the viewing users, and the one or more additional attributes associated with the subject user.