Predicting characteristics of users of a third party system that communicates with an online system and determining accuracy of the predicted characteristics

A method comprising:

• accessing stored information describing a plurality of users of an online system, the information including, for each user of the online system, an asserted characteristic of the user of the online system and a set of other users of the plurality of users of the online system with whom the user of the online system has established a connection in the online system;
• obtaining characteristics of users of a third party system that communicates with the online system, the third party system not maintaining a characteristic of users of the third party system corresponding to the asserted characteristic maintained by the online system;
• identifying a set of users of the online system wherein characteristics of the set of users have at least a threshold similarity to the obtained characteristics of users of the third party system;
• generating a model to determine a predicted characteristic of users of the third party system based on characteristics of users of the set of users of the online system and the asserted characteristic included the stored information describing the plurality of users of an online system, wherein the predicted characteristic of users of the third party system is different from the obtained characteristics of the users of the third party system;
• determining the predicted characteristic of users of the third party system by applying the generated model to characteristics of users of the third party system;
• for each user of at least a group of users of the third party system, determining a distribution of an additional characteristic associated with the user of the third party system;
• clustering at least the group of users of the third party system into one of a plurality of clusters based on the determined distributions;
• for each of the plurality of clusters,
• receiving information from a trusted additional third party system describing a verified characteristic of each of at least a subset of the users of the third party system of the cluster,
• determining an accuracy of the predicted characteristic of the subset of users of the third party system of the cluster by comparing the predicted characteristic of various users of the third party system in the subset to the verified characteristic of various users of the third party system in the subset of users, and
• storing the determined accuracy in association with the cluster;
• receiving a request to present a content item to a viewing user of the third party system based at least in part on a value of the predicted characteristic of the viewing user of the third party system;
• identifying the cluster to which the viewing user of the third party system belongs;
• and
• including the content item in one or more selection processes selecting content for presentation to the viewing user of the third party system based at least in part on the predicted characteristic of the viewing user and the accuracy of the cluster to which the viewing user of the third party system belongs.

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:

• access stored information describing a plurality of users of an online system, the information including, for each user of the online system, an asserted age of the user of the online system and a set of other users of the plurality of users of the online system with whom the user of the online system has established a connection in the online system;
• obtain characteristics of users of a third party system that communicates with the online system, the third party system not maintaining stored information including asserted ages of users of the third party system;
• identify a set of users of the online system so characteristics of the set of users have at least a threshold similarity to the obtained characteristics of users of the third party system;
• generate a model to determine a predicted age of users of the third party system based on characteristics of users of the set of users of the online system and asserted ages of users of the set of users of the online system;
• determine predicted ages of users of the third party system by applying the generated model to characteristics of users of the third party system;
• for each user of at least a group of users of the third party system, determine a distribution of predicted ages of other users of the third party system having a connection to the user of the third party system via the third party system;
• cluster at least the group of users of the third party system into one of a plurality of clusters based on the determined distributions;
• for each of the plurality of clusters,
• receive information from a trusted additional third party system describing a verified age of each of at least a subset of the users of the third party system in the cluster,
• determine an accuracy of the predicted ages of the subset of users of the third party system in the cluster by comparing the predicted ages to the verified ages of the subset of users of the third party system in the cluster, and
• store the determined accuracy in association with the cluster;
• receive a request to present a content item to a viewing user of the third party system based at least in part on an age of a viewing user;
• identify the cluster to which the viewing user of the third party system belongs;
• and
• include the content item in one or more selection processes selecting content for presentation to the viewing user of the third party system based at least in part on a predicted age of the viewing user of the third party system and the accuracy of the cluster including the viewing user of the third party system.