Abstract: Disclosed is an online system that infers interests of unresolved users for whom the interests are not known. The online system determines certain features about the unresolved users, but does not have certain information about the users themselves (e.g., their interests), so instead infers these attributes based on the features of the user. The online system provides the features as input to a classifier trained to predict a particular interest, and the classifier outputs a prediction of whether the user has the corresponding interest. In one embodiment, the online system trains a classifier for various interest values by forming training sets for the interests using the features for users who are logged into the online system and hence have known interests.

Abstract: An online system maintains characteristics for its users and may access characteristics of users maintained by a third party system. The online system may select content for a user of the third party system based on characteristics maintained by the third party system. If the third party system does not maintain a characteristic for its users, the generates a model predicting the characteristic for third party system users based on a set of online system users identified based on characteristics of third party system users. The online system clusters third party system users based on the predicted characteristic for other third party system users connected to the third party system user. Using verified characteristics for third party system users from a trusted third party system, the online system determines an accuracy of the predicted characteristic for third party system users in a cluster.

Abstract: An online system obtains a set of resolved impressions based on historical data about multiple publishers. A set of features is then extracted, for each resolved impression, based on a comparison of historical data about the first publisher and the second publisher. The online system performs training of a machine-learned model based on the set of features. Data about a plurality of new impressions are input into the trained machine-learned model to obtain an output of the trained machine-learned model. A reach overlap metric and unique reach metric can be computed based on the output of the trained machine-learned model.