Systems, methods, and storage media for training a machine learning model

Abstract

Systems, methods, and storage media for training a machine learning model are disclosed. Exemplary implementations may select a set of training images for a machine learning model, extract object features from each training image to generate an object tensor for each training image, extract stylistic features from each training image to generate a stylistic feature tensor for each training image, determine an engagement metric for each training image, and train a neural network comprising a plurality of nodes arranged in a plurality of sequential layers.

What is claimed is: 
     
         1 . A method, comprising:
 accessing, by one or more processors, a web-based property over a network, the web-based property containing a plurality of images;   extracting, by the one or more processors, an image and image metadata associated with the image from the web-based property;   determining, by the one or more processors, a target audience for the web-based property;   identifying, by the one or more processors, a training data set that corresponds to the determined target audience;   aggregating, by the one or more processors, the image and the image metadata into the training data set based on the target audience for the web-based property; and   training, by the one or more processors, a machine learning model with the training data set with the aggregated image and image metadata.   
     
     
         2 . The method of  claim 1 , wherein determining the target audience for the web-based property comprises:
 identifying, by the one or more processors, demographic, psychographic, or behavioral characteristics of users that visit the set of web-based properties; and   determining, by the one or more processors, the target audience based on the identified demographic, psychographic, or behavioral characteristics.   
     
     
         3 . The method of  claim 2 , further comprising:
 identifying, by the one or more processors, a set of web-based properties that correspond to the target audience for the web-based properties of the set.   
     
     
         4 . The method of  claim 3 , wherein determining the target audience based on the demographic, psychographic, or behavioral characteristics comprises determining, by the one or more processors, the target audience for the web-based property based on the web-based property being a web-based property of the identified set of web-based properties. 
     
     
         5 . The method of  claim 3 , further comprising ranking, by the one or more processors, the set of web-based properties with rankings based on proportions of user accounts that visit or engage with the web-based properties of the set of web-based properties and that are members of the target audience. 
     
     
         6 . The method of  claim 5 , further comprising:
 identifying, by the one or more processors, a subset of the set of web-based properties based on the rankings of the set of web-based properties,   wherein determining the target audience based on the demographic, psychographic, or behavioral characteristics comprises determining, by the one or more processors, the target audience for the web-based property based on the web-based property being a web-based property of the identified subset of web-based properties.   
     
     
         7 . The method of  claim 6 , wherein extracting the image and the image metadata from the web-based property is performed in response to determining the web-based property is a web-based property of the subset of the set of web-based properties. 
     
     
         8 . The method of  claim 3 , further comprising:
 determining, by the one or more processors, a respective audience relevance metric for each web-based property of the set of web-based properties; and   identifying, by the one or more processors, a subset of the set of web-based properties based on the respective audience relevance metrics.   
     
     
         9 . The method of  claim 8 , wherein determining the respective audience relevance metric for each web-based property comprises determining, by the one or more processors, a respective audience relevance metric for a second web-based property based on a number of images that are posted to the second web-based property, a number of visitors to the second web-based property, an amount of engagement that posted images receive on the second web-based property, or a quality of images that are posted to the second web-based property. 
     
     
         10 . The method of  claim 8 , wherein determining the respective audience relevance metric for each web-based property comprises determining a respective audience relevance metric for a second web-based property based on a product category for the second web-based property. 
     
     
         11 . The method of  claim 1 , further comprising:
 selecting, by the one or more processors, the image from the plurality of images of the web-based property based on the image metadata and an image metadata qualification criteria.   
     
     
         12 . The method of  claim 11 , wherein the image metadata qualification criteria comprises a minimum number of user interactions or a minimum number of comments. 
     
     
         13 . The method of  claim 1 , further comprising removing, by the one or more processors, images extracted from a second web-based property from the training data set in response to determining the second web-based property has a number of posted images below a first threshold or a number of followers below a second threshold. 
     
     
         14 . The method of  claim 1 , further comprising:
 selecting, by the one or more processors, the image from the plurality of images of the web-based property based on a product, an image subject matter, or a scene depicted in the image and an image content qualification criteria.   
     
     
         15 . The method of  claim 1 , further comprising:
 executing, by the one or more processors, the trained machine learning model using a second image and second image metadata associated with the second image, the executing causing the trained machine learning model to output a performance score for the second image.   
     
     
         16 . A system, the system comprising:
 one or more hardware processors configured by machine-readable instructions to:
 access a web-based property over a network, the web-based property containing a plurality of images; 
 extract an image and image metadata associated with the image from the web-based property; 
 determine a target audience for the web-based property; 
 identify a training data set that corresponds to the determined target audience; 
 aggregate the image and the image metadata into the training data set based on the target audience for the web-based property; and 
 train a machine learning model with the training data set with the aggregated image and image metadata. 
   
     
     
         17 . The system of  claim 16 , wherein the one or more hardware processors are configured to determine the target audience for the web-based property by:
 identifying demographic, psychographic, or behavioral characteristics of users that visit the set of web-based properties; and   determining the target audience based on the identified demographic, psychographic, or behavioral characteristics.   
     
     
         18 . The system of  claim 16 , wherein the one or more hardware processors are further configured to:
 select the image from the plurality of images of the web-based property based on a content of the image, the image metadata, and an image metadata qualification criteria.   
     
     
         19 . A non-transitory computer-readable storage medium having instructions embodied thereon, the instructions being executable by one or more processors to perform a method, the method comprising:
 accessing a web-based property over a network, the web-based property containing a plurality of images;   extracting an image and image metadata associated with the image from the web-based property;   determining a target audience for the web-based property;   identifying a training data set that corresponds to the determined target audience;   aggregating the image and the image metadata into the training data set based on the target audience for the web-based property; and   training a machine learning model with the training data set with the aggregated image and image metadata.   
     
     
         20 . The non-transitory computer-readable storage medium of  claim 19 , wherein determining the target audience for the web-based property comprises:
 identifying demographic, psychographic, or behavioral characteristics of users that visit the set of web-based properties; and   determining the target audience based on the identified demographic, psychographic, or behavioral characteristics.