Demographic Audience Classification for Cookie-Free Impression Deduplication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for tracking digital media viewership, such as server logs and third-party cookies, are susceptible to over-counting and under-counting due to caching and privacy concerns, leading to inaccurate audience metrics and duplicate impressions across multiple devices.
Innovation Solution
A system utilizing database proprietor subscriber-level audience metrics and machine learning models to determine demographic classifications for census-level impression counts and unique audience sizes, leveraging media tags and identification information without relying on third-party cookies, by encoding features in media tags and training models with streaming meter data to deduplicate impressions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If third-party cookies are used for tracking digital media viewership, then impression counts can be collected across multiple devices, but over-counting and under-counting occur due to caching and privacy concerns
Solution Approach 1:
The patent introduces an intermediary deduplication system that sits between data collection and audience metrics generation. This system uses probabilistic data structures (Bloom filters, hyperloglog) as intermediaries to process and deduplicate impression data from multiple sources before generating final audience metrics, thereby resolving the over-counting and under-counting issues without relying on third-party cookies
Solution Approach 2:
The patent replaces the mechanical cookie-based tracking system with a probabilistic computational approach. Instead of relying on persistent identifiers like cookies that are subject to caching and privacy issues, the system uses probabilistic data structures and machine learning models to infer unique audience sizes and demographic classifications, achieving more reliable measurement
2Productivity
If server logs are used to track digital media access, then viewership information can be obtained, but duplicate impressions across multiple devices are not eliminated
Solution Approach 1:
The patent applies preliminary deduplication actions by processing impression data through probabilistic data structures before final analysis. By pre-processing the data to estimate unique audience sizes using hyperloglog and bloom filters, the system eliminates duplicate impressions across devices before generating audience metrics, improving measurement precision while maintaining tracking efficiency
Solution Approach 2:
The patent creates probabilistic copies of impression data structures (bloom filters, hyperloglog sketches) that can be replicated and merged across distributed systems. These probabilistic copies allow efficient deduplication of impressions across multiple devices without requiring centralized storage of all raw impression data, maintaining both efficiency and accuracy
3Measurement precision
If census-level impression data is collected without demographic information, then complete impression counts are obtained, but demographic classifications cannot be determined
Solution Approach 1:
The patent implements feedback loops where probabilistic demographic estimates are continuously refined using aggregated census-level data. The system collects complete impression data without demographic filters, generates initial demographic classifications using probabilistic models, then uses feedback from aggregated results to refine and improve demographic accuracy over time, maintaining both completeness and information quality
4Measurement precision
If machine learning models are trained with streaming meter data, then demographic classifications can be determined, but computational complexity increases
Solution Approach 1:
The patent segments the demographic classification process into multiple stages: initial probabilistic classification using lightweight features, then iterative refinement using machine learning models on aggregated data. This segmentation allows the system to achieve high demographic accuracy while managing computational complexity by processing data in manageable stages rather than requiring all computations simultaneously
Data Source
AI summary
Methods, apparatus, systems and articles of manufacture to determine demographic classifications for census level impression counts and unique audience sizes are disclosed. In an example, the apparatus includes media tag format circuitry to generate a reformatted media tag corresponding to an impression request. The example apparatus also includes model execution circuitry to execute a machine learning model based on the reformatted media tag to generate outputs, the outputs including at least a value representative of a probability of an occurrence of a demographic classification. The example apparatus further includes audience counting circuitry to assign an identification of ones of audience members in a group to the demographic classification based at least on the outputs.