Iterative Classifier Training on Online Social Networks

Abstract

One or more computing devices may access, from a set of profile objects each associated with an entity profile, a set of training objects each associated with an object-classification. The devices may identify, by an object-classifier algorithm, from a first set of content objects associated with the set of training objects, a first set of features each having a feature score indicating a correlation value with the object-classification. Based on the first set of features the devices may update the object-classifier algorithm and classify, by the updated object-classifier algorithm, one or more of the profile objects as being associated with the object-classification. The devices may identify, by the updated object-classifier algorithm, from a second set of content objects associated with the classified profile objects, a second set of features having feature scores greater than a threshold feature score, and update the object-classifier algorithm based on the second set of features.

What is claimed is: 
     
         1 . A method comprising, by one or more computing devices:
 accessing, from a set of profile objects each associated with an entity profile, a set of training objects each associated with an object-classification;   identifying, by an object-classifier algorithm, from a first set of content objects associated with the set of training objects, a first set of features each having a feature score indicating a correlation value with the object-classification;   updating the object-classifier algorithm based on the first set of features;   classifying, by the updated object-classifier algorithm, one or more of the profile objects as being associated with the object-classification;   identifying, by the updated object-classifier algorithm, from a second set of content objects associated with the classified profile objects, a second set of features having feature scores greater than a threshold feature score; and   updating the object-classifier algorithm based on the second set of features.   
     
     
         2 . The method of  claim 1 , further comprising:
 classifying, based on the updated object-classifier algorithm based on the second set of features, one or more profile objects associated with one or more respective users of an online network to determine, for each user, an interest of the user in a particular activity or service.   
     
     
         3 . The method of  claim 1 , wherein updating the object-classifier algorithm based on the second set of features comprises adding, to the object-classifier algorithm, one or more of the second set of features having feature scores greater than a predetermined threshold feature score. 
     
     
         4 . The method of  claim 3 , further comprising:
 removing, from the object-classifier algorithm, one or more of the first set of features having a feature score less than the predetermined threshold feature score.   
     
     
         5 . The method of  claim 4 , wherein the predetermined threshold feature score is based on a percentile of the feature scores of the second set of features. 
     
     
         6 . The method of  claim 5 , wherein the predetermined threshold feature score is greater than 90% of the feature scores of the second set of features. 
     
     
         7 . The method of  claim 1 , wherein one or more content objects from the first or second sets of content objects comprise one or more of photos, posts, pages, applications, events, locations, or user groups associated with the entity profile. 
     
     
         8 . The method of  claim 7 , wherein one or more content objects from the first or second sets of content objects are associated with an online social network. 
     
     
         9 . The method of  claim 1 , wherein one or more profile objects of the set of profile objects are associated with one or more posts on a third-party website. 
     
     
         10 . The method of  claim 1 , wherein classifying the one or more of the profile objects as being associated with the object-classification comprises comparing each profile object to a set of filtering criteria. 
     
     
         11 . The method of  claim 10 , wherein the set of filtering criteria is based on metadata associated with the profile objects and the content objects associated with the profile objects. 
     
     
         12 . The method of  claim 11 , wherein the metadata comprises one or more of:
 a number of associated comments;   inclusion or exclusion of one or more text strings;   inclusion or exclusion of one or more punctuation marks;   an association with one or more object-types;   an association with one or more languages;   an association with one or more places, entities, or resource-types;   an association with one or more other objects; or   an association with a particular period of time.   
     
     
         13 . The method of  claim 12 , wherein the one or more punctuation marks comprise one or more of a question mark or exclamation mark. 
     
     
         14 . The method of  claim 1 , wherein training the object-classifier algorithm is further based on the object-classifier algorithm trained in a prior iteration of an iterative training process. 
     
     
         15 . The method of  claim 1 , further comprising:
 accessing an entity graph comprising a plurality of nodes and a plurality of edges connecting the nodes, each node corresponding to a profile object of the set of profile objects each associated with the entity profile.   
     
     
         16 . One or more computer-readable non-transitory storage media embodying software that is operable when executed to:
 access, from a set of profile objects each associated with an entity profile, a set of training objects each associated with an object-classification;   identify, by an object-classifier algorithm, from a first set of content objects associated with the set of training objects, a first set of features each having a feature score indicating a correlation value with the object-classification;   update the object-classifier algorithm based on the first set of features;   classify, by the updated object-classifier algorithm, one or more of the profile objects as being associated with the object-classification;   identify, by the updated object-classifier algorithm, from a second set of content objects associated with the classified profile objects, a second set of features having feature scores greater than a threshold feature score; and   update the object-classifier algorithm based on the second set of features.   
     
     
         17 . The media of  claim 16 , wherein the software is further operable when executed to:
 classify, based on the updated object-classifier algorithm based on the second set of features, one or more profile objects associated with one or more respective users of an online network to determine, for each user, an interest of the user in a particular activity or service.   
     
     
         18 . A system comprising: one or more processors; and a memory coupled to the processors comprising instructions executable by the processors, the processors operable when executing the instructions to:
 access, from a set of profile objects each associated with an entity profile, a set of training objects each associated with an object-classification;   identify, by an object-classifier algorithm, from a first set of content objects associated with the set of training objects, a first set of features each having a feature score indicating a correlation value with the object-classification;   update the object-classifier algorithm based on the first set of features;   classify, by the updated object-classifier algorithm, one or more of the profile objects as being associated with the object-classification;   identify, by the updated object-classifier algorithm, from a second set of content objects associated with the classified profile objects, a second set of features having feature scores greater than a threshold feature score; and   update the object-classifier algorithm based on the second set of features.   
     
     
         19 . The system of  claim 18 , wherein the processors are further operable when executing the instructions to:
 classify, based on the updated object-classifier algorithm based on the second set of features, one or more profile objects associated with one or more respective users of an online network to determine, for each user, an interest of the user in a particular activity or service.   
     
     
         20 . The system of  claim 18 , wherein one or more content objects from the first or second sets of content objects comprise one or more of photos, posts, pages, applications, events, locations, or user groups associated with the entity profile.