Selecting one or more components 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:

• identify an opportunity to present content to a subject user of an online system, the subject user associated with one or more attributes;
• identify a plurality of candidate content item components eligible to be included in an optimal content item for presentation to the subject user, each of the plurality of candidate content item components associated with a set of features;
• retrieve a training set of affinity scores for each of the plurality of candidate content item components, the affinity scores based at least in part on a value of a performance metric associated with an impression of a training content item including a first set of the plurality of candidate content item components previously presented to one or more viewing users of the online system, each of the one or more viewing users associated with one or more additional attributes;
• predict, using a machine-learned model trained on the training set of affinity scores, an affinity score of the subject user for each of the plurality of candidate content item components based at least in part on, the one or more attributes associated with the subject user and the set of features associated with each of the plurality of candidate content item components;
• select a second set of the plurality of candidate content item components based at least in part on the affinity score of the subject user predicted for each of the plurality of candidate content item components, the second set of the plurality of candidate content item components associated with a predicted marginal effect on the value of the performance metric associated with an impression of the optimal content item including one or more of the candidate content item components in the second set;
• generate a plurality of candidate content items, each of the plurality of candidate content items comprising a subset of the second set of the plurality of candidate content item components, each subset different from the others;
• compute a total affinity score of the subject user for each of the plurality of candidate content items based at least in part on a weight associated with a type of each of the candidate content item components in each candidate content item;
• rank the plurality of candidate content items based in part on the total affinity scores;
• select an optimal content item from the one or more of the plurality of candidate content items based at least in part on the ranking;
• and
• present the optimal content item to the subject user.

A method comprising:

• identifying an opportunity to present content to a subject user of an online system, the subject user associated with one or more attributes;
• for different sets of a plurality of candidate content item components eligible to be included in an optimal content item for presentation to the subject user, predicting, using a machine-learned model, a value of a performance metric associated with potential impressions of a content item including each of the different sets of the candidate content item components presented to the subject user, the value predicted based at least in part on the one or more attributes associated with the subject user and one or more features associated with the different sets of the candidate content item components;
• selecting one or more of the plurality of candidate content item components based at least in part on a predicted marginal effect the value of the performance metric associated with potential impressions of the content item including each of the different sets of the candidate content item components;
• generating a plurality of candidate content items, each of the plurality of candidate content items comprising one or more of the plurality of candidate content item components, each candidate content item comprising different candidate content item components;
• selecting an optimal content item from the one or more of the plurality of candidate content items based at least in part on a ranking of a total affinity score of the subject user for each of the plurality of candidate content items, the total affinity score computed based at least in part on a weight associated with a type of the candidate content item components in each candidate content item;
• and
• presenting the optimal content item to the subject user.