Machine-Learning Cell Twinning for Unbiased Content Measurement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face errors in determining digital content performance due to differences in user profiles and behaviors between control and exposed groups, leading to biased results and resource-intensive bias correction methods.
Innovation Solution
Utilize machine learning to generate a parameter ranking and create a digital twin of computing devices that did not access the digital content, ensuring similarity in parameters such as demographics and behavior, reducing bias and resource consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional control groups are used to determine digital content performance, then performance measurement is possible, but user profile and behavior differences cause biased results
Solution Approach 1:
The patent creates a digital twin of the control group by copying and adapting the exposed group's parameters. Instead of using a traditional control group that may differ in user profiles and behaviors, the system generates a synthetic control group that mirrors the exposed group's characteristics while varying only the content exposure variable, thereby eliminating selection bias and improving measurement reliability
Solution Approach 2:
The system systematically varies specific parameters (content exposure) while holding other parameters constant through the digital twin construction. By controlling and standardizing parameters such as user demographics, device characteristics, and viewing conditions between the exposed and control groups, the patent isolates the effect of content exposure and improves measurement precision
2Measurement precision
If bias correction methods are applied to traditional control groups, then result accuracy improves, but processing resources and power consumption increase
Solution Approach 1:
The patent performs bias correction in advance by constructing the control group digital twin before the performance measurement experiment. By pre-aligning the parameter distributions of the exposed and control groups through machine learning-based parameter ranking and matching, the system eliminates the need for complex post-hoc statistical bias correction methods, reducing computational resources and power consumption during the actual measurement phase
3Reliability
If detailed parameter matching is performed between control and exposed groups, then bias reduction improves, but system complexity increases
Solution Approach 1:
The system employs machine learning models that automatically perform parameter ranking, weighting, and matching without requiring manual configuration or expert intervention. The algorithm autonomously identifies which parameters are most important for matching and adjusts the control group composition accordingly, simplifying the system architecture while achieving comprehensive parameter alignment and reducing bias
Data Source
AI summary
Twinning is disclosed based on parameter ranking. A system can include a data processing system. The data processing system can include one or more processors, couple with memory. The data processing system can receive identifiers of computing devices of a cell that accessed digital content. The data processing system can receive parameters linked with the identifiers. The data processing system can determine, based on the parameters and a model trained with machine learning, a ranking of the parameters. The data processing system can generate, based on the ranking of the parameters, a twin of the cell comprising identifiers of computing devices that did not access the digital content.