US 12,462,266 B2
Methods and systems for evaluating content
James Gin, London (GB); Chris Loy, London (GB); and Igor Volzhanin, London (GB)
Assigned to SHUTTERSTOCK.AI INTERNATIONAL LIMITED, London (GB)
Filed by Datasine Limited, London (GB)
Filed on Dec. 17, 2021, as Appl. No. 17/555,100.
Claims priority of application No. 2020206 (GB), filed on Dec. 18, 2020.
Prior Publication US 2022/0198486 A1, Jun. 23, 2022
Int. Cl. G06Q 30/0202 (2023.01); G06Q 30/0201 (2023.01)
CPC G06Q 30/0202 (2013.01) [G06Q 30/0201 (2013.01)] 14 Claims
OG exemplary drawing
 
1. A computer-implemented method for automatically analyzing content to predict a consumer interaction rate of said content, based on historical consumer interaction rates of historical content, the method comprising:
receiving content data from a computing platform, the content data relating to a first content displayed via a digital platform and comprising image data and/or text data, the content data having associated contextual metadata defining contextual attributes of said first content;
processing the image data and/or the text data to determine feature vectors of the image data and/or the text data;
inputting the associated contextual metadata and the feature vectors of the image data and/or the text data into a first machine learning (ML) model;
predicting, using the first ML model, where said first content would be ranked within a plurality of historical content, the plurality of historical content having similar associated contextual metadata to the said first content, wherein the plurality of historical content is ranked by a historical consumer interaction rate, and wherein the historical consumer interaction rate is based on a plurality of types of actions taken by a consumer in response to the plurality of historical content;
outputting a prior probability from the first ML model, the prior probability being based on a predicted rank of said first content;
receiving, from the computing platform, additional metadata relating to said first content;
inputting the additional metadata and the prior probability into a second ML model to output a first posterior probability representing a first predicted consumer interaction rate of said first content;
retrieving, from the computing platform, impression data relating to a first performance of said first content;
updating the second ML model based on the impression data;
receiving, from the second ML model, a second posterior probability representing a second predicted consumer interaction rate of said first content based on the impression data; and
updating, based on the second posterior probability, the digital platform by replacing the first content with a second content having a higher predicted consumer interaction rate.