Expansion of targeting criteria based on advertisement performance

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

• receive an advertisement request including targeting criteria specifying a target audience of users of an online system eligible to be presented with the advertisement request, the advertisement request specifying a goal and a time interval for achieving the goal;
• determine an actual performance of the advertisement request during a time period within the time interval based on presentation of advertisement content included in the advertisement request to users of an online system;
• compare the actual performance of the advertisement request to a portion of the goal associated with the time period;
• and
• responsive to the comparison determining that the actual performance of the advertisement request does not satisfy the portion of the goal associated with the time period:
• identify a cluster group that is associated with the targeting criteria included in the advertisement request;
• select a sample set of users for determining a cluster cutoff score corresponding to a target ranking threshold based on the goal;
• determine a set of sampling scores for the sample set of users;
• determine a ranking of the sample set of users based on the respective sampling scores;
• select a subset of the sample scores above a target ranking threshold and within a threshold amount of the target ranking threshold;
• perform a data fit between the subset of the sample scores and the ranking to estimate a cluster cutoff score corresponding to the target ranking threshold, wherein performing the data fit to estimate the cluster cutoff score comprises:
• performing the data fit on a plurality of computational nodes to generate a plurality of results;
• sharing the results between neighboring computational nodes;
• and
• selecting the cluster cutoff score in response to a consensus number of computational nodes agreeing on the cluster cutoff score;
• identify a target user outside of the target audience that has user characteristics failing to satisfy the targeting criteria;
• determine, based on a cluster model, a cluster score for the user representing an affinity measure between characteristics and the targeting criteria for the cluster group;
• compare the cluster score to the cluster cutoff score to determine if the target user is a member of the cluster group;
• and
• responsive to determining that the target user is a member of the cluster group, add the target user to the target audience eligible to be presented with the advertisement request.