Bloom Filter Cardinality Estimation for Privacy-Safe Audience Deduplication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audience measurement systems face challenges in accurately estimating unique audience size across multiple datasets while preserving user privacy, as they rely on third-party cookies which may be limited or unavailable, and deduplication techniques involving personally identifiable information (PII) are undesirable due to privacy concerns and computational inefficiencies.
Innovation Solution
The use of Bloom filter arrays to generate sketch data that summarizes media exposure across multiple datasets, allowing for the estimation of cardinality without revealing individual identities, thereby enabling accurate deduplication of audience members and preserving privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deduplication techniques using personally identifiable information (PII) are employed to estimate unique audience size, then measurement precision is improved, but privacy protection deteriorates and device complexity increases
Solution Approach 1:
The patent extracts only the essential identifying features needed for deduplication while removing personally identifiable information. Bloom filters capture presence/absence patterns of users across datasets without storing actual PII, thereby achieving deduplication functionality while protecting user privacy.
Solution Approach 2:
Bloom filters serve as an intermediary data structure between raw PII and final audience metrics. They transform detailed user information into a compact probabilistic representation that enables deduplication across datasets while preventing direct access to individual user identities.
2Measurement precision
If traditional deduplication methods using PII are used, then cardinality estimation accuracy is improved, but computational efficiency deteriorates and memory usage increases
Solution Approach 1:
The patent creates a compact copy of user presence information using Bloom filters instead of storing complete user profiles. This copied representation maintains sufficient information for deduplication while dramatically reducing memory requirements and computational overhead for processing large datasets.
Solution Approach 2:
The patent changes the parameter representation from detailed PII fields to a compact bit array structure. By transforming user information into a fixed-size Bloom filter representation, the system achieves constant-time operations regardless of dataset size, improving computational efficiency while maintaining estimation accuracy.
3Measurement precision
If complete user data is shared across multiple database proprietors for audience measurement, then measurement precision is improved, but privacy protection deteriorates
Solution Approach 1:
The patent extracts only the minimal necessary information for cross-platform deduplication - specifically, user presence patterns across datasets - while leaving out all personally identifiable information. This extracted representation enables accurate audience measurement across multiple database proprietors without exposing user privacy.
Solution Approach 2:
Bloom filters act as a privacy-preserving intermediary that allows multiple database proprietors to collaboratively measure cross-platform audiences. Each proprietor can contribute to the aggregated measurement using their own Bloom filters without sharing raw user data, enabling precision while maintaining privacy boundaries.
Data Source
AI summary
Methods and apparatus to estimate cardinality across multiple datasets represented using Bloom filter arrays are disclosed. Disclosed examples include processor circuitry to execute and/or instantiate instructions to determine an inclusion-exclusion expression that defines an audience size for a user group of interest. Terms in the inclusion-exclusion expression corresponding to either a first cardinality of a first one of at least three Bloom filter arrays or a second cardinality of a union of two or more of the Bloom filter arrays. Different ones of the Bloom filter arrays representative of different sets of users who accessed media. The at least one processor further to estimate, based on the inclusion-exclusion expression, the audience size of the user group of interest.