Audience on networked devices

Abstract

A method, system, and apparatus provide the ability to target an audience of network devices. One of the methods includes receiving information from a source, the information associated with a device identifier. The method includes determining, based on the device identifier, a unique user identifier, wherein the unique user identifier identifies a user independent of network, media, and location. The method includes identifying at least one user attribute based on the received information. The method includes associating the user attribute with the unique user identifier. The method also includes storing the user attribute in a repository. The repository stores a plurality of other user attributes associated with the unique user identifier, the other stored user attributes being received from a plurality of different sources.

What is claimed is: 
     
         1 . A computer-implemented method, comprising:
 (a) receiving information from a source, the information associated with a device identifier, wherein the source corresponds to a networked device;   (b) determining, based on the device identifier, a unique user identifier, wherein the unique user identifier identifies a user independent of network, media, and location;   (c) identifying at least one user attribute for the user based on the received information;   (d) determining an inferred user attribute for the user using an inference model, wherein the user is without the inferred user attribute, and wherein the determining comprises:
 (1) obtaining one or more prediction attributes, wherein the one or more prediction attributes are attributes common to a set of users of whom the inferred user attribute will be derived; 
 (2) utilizing a ranking function to generate a lift curve on the one or more prediction attributes; 
 (3) generating, using the inference model, a ranking score for the user; 
 (4) determining that the ranking score is above a certain level of lift of the lift curve; 
 (5) utilizing the inference model to determine the inferred user attribute from the one or more prediction attributes; 
   (e) associating the user attribute and the inferred user attribute with the unique user identifier;   (f) storing the user attribute and inferred user attribute in a repository, wherein the repository stores a plurality of other user attributes associated with the unique user identifier, the other stored user attributes being received from a plurality of different sources, wherein the plurality of different sources correspond to a plurality of different devices; and   (g) targeting the user at one or more of the different devices based on the user attribute and inferred user attribute.   
     
     
         2 . The computer-implemented method of  claim 1 , wherein:
 the inference model splits the one or more prediction attributes into a training set and a test set;   the inference model utilizes the training set to output the ranking function;   the lift curve is generated based on the ranking function and the test set;   the lift curve indicates, based on a threshold level, whether results of the inference model should be used for the user.   
     
     
         3 . The computer-implemented method of  claim 1 , wherein:
 the lift curve is generated using cross validation and a bootstrap.   
     
     
         4 . The computer-implemented method of  claim 1 , wherein:
 the inferred user attribute is determined in real time with the model.   
     
     
         5 . The computer-implemented method of  claim 1 , wherein:
 the inference model has a low prediction accuracy;   the lift curve enables the inference model to do inference confidence.   
     
     
         6 . The computer-implemented method of  claim 1 , further comprising:
 building behavior interest scores in context categories comprising:
 associating one or more intention scores with product and service categories; 
 associating a propensity of action for a user event with ad event attributes; and 
 associating lift time value scores with specific business categories. 
   
     
     
         7 . The computer-implemented method of  claim 1 , further comprising:
 receiving additional information about the user from a plurality of sources, each source associated with a device identifier; and   associating each of the device identifiers with the unique user identifier.   
     
     
         8 . The computer-implemented method of  claim 1 , further comprising:
 receiving a request for user attributes, the request including a device identifier;   identifying a unique user identifier based on the device identifier; and   providing user data associated with the unique user identifier.   
     
     
         9 . The computer-implemented method of  claim 1 , further comprising:
 targeting a group of users across different device identifiers and media channels comprising:
 receiving request that contains a linked device id or user signature; 
 receiving a rule involving targeted user attributes; 
 retrieving the stored user attribute and inferred user attribute using a user device id or a user signature; and 
 identifying the targeted group of users by applying the rule to the user attributes and the inferred user attribute. 
   
     
     
         10 . The computer-implemented method of  claim 1 , wherein:
 the targeting comprises targeting the user from the same device or retargeting the user from a different device based on the ranking score.   
     
     
         11 . A system comprising:
 (a) one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:
 (1) receiving information from a source, the information associated with a device identifier, wherein the source corresponds to a networked device; 
 (2) determining, based on the device identifier, a unique user identifier, wherein the unique user identifier identifies a user independent of network, media, and location; 
 (3) identifying at least one user attribute for the user based on the received information; 
 (4) determining an inferred user attribute for the user using an inference model, wherein the user is without the inferred user attribute, and wherein the determining comprises:
 (i) obtaining one or more prediction attributes, wherein the one or more prediction attributes are attributes common to a set of users of whom the inferred user attribute will be derived; 
 (ii) utilizing a ranking function to generate a lift curve on the one or more prediction attributes; 
 (iii) generating, using the inference model, a ranking score for the user; 
 (iv) determining that the ranking score is above a certain level of lift of the lift curve; 
 (v) utilizing the inference model to determine the inferred user attribute from the one or more prediction attributes; 
 
 (5) associating the user attribute and the inferred user attribute with the unique user identifier; 
 (6) storing the user attribute and inferred user attribute in a repository, wherein the repository stores a plurality of other user attributes associated with the unique user identifier, the other stored user attributes being received from a plurality of different sources, wherein the plurality of different sources correspond to a plurality of different devices; and 
 (7) targeting the user at one or more of the different devices based on the user attribute and inferred user attribute. 
   
     
     
         12 . The system of  claim 11 , wherein:
 the inference model splits the one or more prediction attributes into a training set and a test set;   the inference model utilizes the training set to output the ranking function;   the lift curve is generated based on the ranking function and the test set;   the lift curve indicates, based on a threshold level, whether results of the inference model should be used for the user.   
     
     
         13 . The system of  claim 11 , wherein:
 the lift curve is generated using cross validation and a bootstrap.   
     
     
         14 . The system of  claim 11 , wherein:
 the inferred user attribute is determined in real time with the model.   
     
     
         15 . The system of  claim 11 , wherein:
 the inference model has a low prediction accuracy;   the lift curve enables the inference model to do inference confidence.   
     
     
         16 . The system of  claim 11 , wherein the operations further comprise:
 building behavior interest scores in context categories comprising:
 associating one or more intention scores with product and service categories; 
 associating a propensity of action for a user event with ad event attributes; and 
 associating lift time value scores with specific business categories. 
   
     
     
         17 . The system of  claim 11 , wherein the operations further comprise:
 receiving additional information about the user from a plurality of sources, each source associated with a device identifier; and   associating each of the device identifiers with the unique user identifier.   
     
     
         18 . The system of  claim 11 , wherein the operations further comprise:
 receiving a request for user attributes, the request including a device identifier;   identifying a unique user identifier based on the device identifier; and   providing user data associated with the unique user identifier.   
     
     
         19 . The system of  claim 11 , wherein the operations further comprise:
 targeting a group of users across different device identifiers and media channels comprising:
 receiving request that contains a linked device id or user signature; 
 receiving a rule involving targeted user attributes; 
 retrieving the stored user attribute and inferred user attribute using a user device id or a user signature; and 
 identifying the targeted group of users by applying the rule to the user attributes and the inferred user attribute. 
   
     
     
         20 . The system of  claim 11 , wherein:
 the targeting comprises targeting the user from the same device or retargeting the user from a different device based on the ranking score.