Demographic Prediction Using Aggregated Audit Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current demographic prediction methods for web requests face challenges due to aggregated and batched data from third-party rating services, which are not suitable for real-time decision-making and lack individual user-level data, leading to inaccurate predictions and reduced in-target rates.
Innovation Solution
A system that uses aggregated demographic audit information from third-party measurement companies to predict demographics for web requests by encoding web request properties, measuring disagreement between audit information, and creating a matrix of probabilities to determine the weight of each property for accurate prediction, while performing quality checks and selecting the most accurate demographic prediction model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If aggregated demographic audit information from third-party rating services is used for demographic prediction, then the system can operate without requiring individual user-level data and avoid privacy issues, but the prediction accuracy and in-target rates decrease due to lack of real-time individual user data
Solution Approach 1:
The patent segments the aggregated demographic audit information into multiple properties (e.g., device type, browser, operating system, screen resolution) and creates separate prediction models for each property. This segmentation allows the system to process aggregated data in a structured manner, measuring disagreement between different audit information sources and weighting them appropriately to improve prediction accuracy while maintaining privacy compliance.
Solution Approach 2:
The patent transforms the aggregated demographic audit information by creating a matrix of probabilities that represents the disagreement between different audit properties. It then uses these transformed parameters to weight the importance of each property in the prediction model, effectively changing the parameter representation from raw aggregated counts to probabilistic disagreement measures that enhance prediction accuracy.
2Measurement precision
If first-party demographic data is collected from user profiles and browsing behavior, then demographic prediction accuracy can be improved, but users may not wish to disclose their demographic data and legal restrictions may apply
Solution Approach 1:
The patent introduces third-party rating services as intermediaries that provide aggregated demographic audit information. Instead of directly collecting sensitive user data, the system uses these intermediary services that already have established relationships with users and can provide demographic information in an aggregated, privacy-preserving manner. This intermediary approach maintains prediction accuracy while avoiding direct privacy violations.
Solution Approach 2:
The patent creates a synthetic representation of demographic data through the matrix of probabilities that copies the essential statistical properties of individual user demographics without revealing actual user identities. The aggregated audit information serves as a copy that preserves demographic patterns while eliminating personally identifiable information, allowing accurate predictions without direct user data collection.
3Reliability
If third-party rating company panel data is used for demographic verification, then independent demographic measurement can be achieved, but the data is aggregated and batched which is not suitable for real-time decision-making
Solution Approach 1:
The patent performs preliminary processing of the aggregated third-party rating data by pre-calculating the matrix of probabilities and disagreement measures for different audit properties. This preliminary action transforms the batched data into a format that can be quickly queried and applied in real-time decision-making, effectively bridging the gap between independent measurement reliability and real-time processing speed.
Solution Approach 2:
The patent creates a dynamic prediction model that can adaptively weight different audit properties based on the matrix of probabilities. The system dynamically selects and combines relevant properties from the aggregated third-party data depending on the specific prediction context, enabling real-time flexibility while maintaining the reliability of independent demographic measurement from rating companies.
Data Source
AI summary
Systems and methods are disclosed for demographic prediction based on aggregated training data. The predictions are based on auditing aggregated data associated with identified properties of web requests. The audited information is based on a batch of prior web requests that have the same property and have been audited by a measurement company.