Machine learning techniques for user group based content distribution

A system comprising:

• one or more processors;
• and
• one or more storage devices storing instructions that, when executed by the one or more processors, cause the one or more processor to perform operations comprising:
• identifying, for each user group of a plurality of user groups that each include multiple users classified as having similar interests, a first set of data comprising a first set of attributes of the user group;
• for each user group of the multiple user groups:
• obtaining first party user identifiers for the multiple users of the user group, wherein the first party user identifier for an individual user identifies the individual user to a first party content provider;
• identifying, for each user in the user group, a second set of data describing activity of the user with respect to content of the first party content provider;
• performing, for each user in the user group, a contextual analysis of the first set of data for the user group and the second set of data for the user, thereby generating one or more labels indicating a category of interest of the user;
• and
• generating one or more training samples for each user in the user group, the generating comprising, for each label of the one or more labels for each user in the user group, combining at least a portion of the first set of data for the user group, the second set of data for the user, and the label;
• generating a training dataset comprising the one or more training samples for each user in each of the multiple user groups;
• training, using the training dataset, to generate a trained machine learning model comprising (1) a base machine learning model configured to generate an encoded representation of the training samples, and (2) one or more task-specific machine learning models configured to receive as input, the encoded representation of the training samples and generate, as output, one or more output labels that each indicate an aggregated interest for the users in the user group;
• receiving, from a client device of a given user, a digital component request, wherein the digital component request includes a given user group identifier of a given user group that includes the given user as a member;
• identifying a set of attributes for the given user group;
• providing the set of attributes for the given user group as input to the machine learning model;
• generating, using the trained machine leaning model, an output label that indicates an interest of the given user of the client device;
• selecting one or more digital components based on the output label of the given user;
• and
• transmitting the one or more digital components to the client device.

A computer readable medium storing instructions that, when executed by one or more data processing apparatus, cause the one or more data processing apparatus to perform operations comprising:

• identifying, for each user group of a plurality of user groups that each include multiple users classified as having similar interests, a first set of data comprising a first set of attributes of the user group;
• for each user group of the multiple user groups:
• obtaining first party user identifiers for the multiple users of the user group, wherein the first party user identifier for an individual user identifies the individual user to a first party content provider;
• identifying, for each user in the user group, a second set of data describing activity of the user with respect to content of the first party content provider;
• performing, for each user in the user group, a contextual analysis of the first set of data for the user group and the second set of data for the user, thereby generating one or more labels indicating a category of interest of the user;
• and
• generating one or more training samples for each user in the user group, the generating comprising, for each label of the one or more labels for each user in the user group, combining at least a portion of the first set of data for the user group, the second set of data for the user, and the label;
• generating a training dataset comprising the one or more training samples for each user in each of the multiple user groups;
• training, using the training dataset, to generate a trained machine learning model comprising (1) a base machine learning model configured to generate an encoded representation of the training samples, and (2) one or more task-specific machine learning models configured to receive as input, the encoded representation of the training samples and generate, as output, one or more output labels that each indicate an aggregated interest for the users in the user group;
• receiving, from a client device of a given user, a digital component request, wherein the digital component request includes a given user group identifier of a given user group that includes the given user as a member;
• identifying a set of attributes for the given user group;
• providing the set of attributes for the given user group as input to the machine learning model;
• generating, using the trained machine leaning model, an output label that indicates an interest of the given user of the client device;
• selecting one or more digital components based on the output label of the given user;
• and
• transmitting the one or more digital components to the client device.