Estimating conversions

A computer-readable medium bearing computer-executable instructions that, when executed by an online service operating on a computing system comprising at least a processor executing the instructions, carries out a method comprising:

• generating a training dataset;
• segmenting the training dataset into a plurality of batches of training data;
• training a machine learning model using the plurality of batches of training data to generate a trained machine learning model configured to predict conversion counts and error rates based on input online behaviors, wherein training the machine learning model includes:
• training, using the plurality of batches of training data, the machine learning model by iteratively randomly disabling a processing node of the machine learning model, training the machine learning model with a respective batch of training data of the plurality of batches of training data while the processing node is disabled, and re-enabling the processing node;
• capturing online behaviors of a plurality of subscribers in response to content;
• for each of the plurality of subscribers, determining whether the subscriber is a measurable subscriber or a non-measurable subscriber, wherein a measurable subscriber is a subscriber whose online behaviors are accessible and can be evaluated;
• for each measurable subscriber:
• determining a first conversion count in response to the content;
• and
• adding the first conversion count to a total conversion count for the content;
• for each non-measurable subscriber:
• processing the captured online behaviors of the non-measurable subscriber to determine, using the trained machine learning model, a second conversion count and an estimated error rate of the second conversion count in response to the content, wherein the trained machine learning model is trained to compute conversion counts and corresponding estimated error rates for measurable subscribers based at least in part on the captured online behavior of the subscribers;
• and
• adding the second conversion count to the total conversion count upon a determination that the estimated error rate falls below a threshold;
• and
• storing the total conversion count.

A computer system, comprising:

• one or more processors;
• and
• a memory, storing program instructions that, when executed by the one or more processors, cause the one or more processors to at least:
• capture online behaviors of a plurality of subscribers of an online service that were exposed to a content, each of the plurality of subscribers exposed to the content being an exposed subscriber, wherein the captured online behaviors include online actions, behaviors, and activities in response to the content;
• train a machine learning model to predict conversion counts and corresponding error rates for subscribers based at least in part on the captured online behaviors of the plurality of subscribers in response to the content, wherein training the machine learning model includes:
• generating a training dataset;
• segmenting the training dataset into a plurality of batches of training data;
• and
• training, using the plurality of batches of training data, the machine learning model by iteratively randomly disabling a processing node of the machine learning model, training the machine learning model with a respective batch of training data of the plurality of batches of training data while the processing node is disabled, and re-enabling the processing node;
• for each exposed subscriber, determining whether the exposed subscriber is a measurable subscriber or a non-measurable subscriber, wherein a measurable subscriber is an exposed subscriber of the online service whose online behaviors outside of interaction with the online service are accessible and can be evaluated;
• for each measurable subscriber:
• determine, using the machine learning model, a first conversion count of the measurable subscriber in response to the content;
• and
• add the first conversion count of the measurable subscriber to a total conversion count for the content;
• for each non-measurable subscriber:
• process, using the machine learning model, the captured online behaviors of the non-measurable subscriber to generate a second conversion count and an estimated error rate of the second conversion count in response to the content;
• add the second conversion count to the total conversion count upon a determination that the estimated error rate is below a threshold;
• and
• store the total conversion count in conjunction with the content.