Biased Loss Function for Predictive Send Time Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional technologies for optimizing send times of electronic communications in B2B scenarios fail to accurately predict both open and click times, leading to inefficiencies and missed opportunities due to treating errors on either side of prediction equally, which can result in sending communications too late and missing time-sensitive information.
Innovation Solution
A machine learning model, such as a neural network, is used to contemporaneously predict both open and click times for electronic communications, with a biased loss function that penalizes overpredicting open times, ensuring earlier predictions and optimizing send times based on these predictions to avoid missing timely engagement opportunities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional models predict open time equally treating errors on either side the same, then the model is simple and easy to train, but the prediction accuracy is insufficient for time-sensitive communications
Solution Approach 1:
The patent changes the loss function parameters from symmetric (treating overprediction and underprediction equally) to asymmetric (penalizing overprediction more heavily). This parameter change in the training objective function enables the model to prioritize prediction accuracy in a way that prevents time-sensitive communications from being sent too late, while maintaining the same basic model architecture.
2Measurement precision
If separate models are used to predict open time and click time, then each model can be optimized for its specific task, but additional computing resources are required
Solution Approach 1:
The patent merges the prediction of open time and click time into a single model that handles both tasks simultaneously. This unified approach allows the model to share computational resources and learning patterns between the two prediction tasks, achieving accurate predictions for both open and click times while reducing the computing resources that would be required if separate models were used.
3Productivity
If communications are sent later to optimize for click time, then content engagement may improve, but time-sensitive information may be missed
Solution Approach 1:
The patent applies preliminary action by using the asymmetric loss function to bias predictions toward earlier times during the training phase. This ensures that the model inherently prioritizes timely delivery of communications, sending messages before the recipient is likely to engage, thereby ensuring time-sensitive information is not missed while still optimizing for engagement effectiveness.
Data Source
AI summary
An improved electronic communication system schedules transmission of electronic communications based on a predicted open time and click time. The open and click times are predicted from a machine learning model that is trained to optimize for both tasks. Additionally, when training the machine learning model, the loss used for adjusting the system to achieve a desired accuracy may be a biased loss determined from a function that penalizes overpredicting the open time. As such, the loss value may be determined by different set of rules depending on whether the predicted time is greater than the actual time or not.