Digital Forecaster Using HyperLogLog to Prevent Overcounting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for estimating audience impressions and capacity require days to a week and often overestimate due to audience overlap and identifier complexities, failing to accurately fulfill advertiser requirements in finite advertisement space.
Innovation Solution
A system using hyperloglog (HLL) data structures to generate target segment data, intersect with audience member subsets, and forecast future impressions, accounting for audience identifier types and probabilistic placements to ensure accurate capacity estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional techniques are used to estimate audience impressions, then measurement precision can be achieved, but the estimation process requires days to a week and produces inaccurate results due to audience overlap
Solution Approach 1:
The system performs preliminary actions by pre-processing audience data into segmented structures and pre-calculating intersection metrics between different audience segments. This allows the actual estimation to be performed much faster while maintaining accuracy, as the heavy computational work is done in advance on prepared data structures.
Solution Approach 2:
The patent segments the audience into distinct groups (e.g., by demographics, viewing behavior, device type) and maintains separate data structures for each segment. This segmentation allows the system to efficiently calculate intersections between segments without processing the entire audience population at once, dramatically reducing computation time while improving precision by accounting for overlapping segments.
2Ease of operation
If audience segments are summed to estimate total impressions, then the estimation process is simple, but the result overestimates the actual unique impressions due to segment overlap
Solution Approach 1:
The system merges multiple audience segment data structures into a unified structure that maintains information about all segments simultaneously. By using combined data structures that track both individual segment members and their intersections, the system can accurately count unique impressions without simply summing segment sizes, thus maintaining simplicity while eliminating overestimation.
Solution Approach 2:
The system incorporates feedback mechanisms that continuously adjust estimation calculations based on actual audience intersection data. By monitoring and using feedback from identified overlaps between segments, the system refines its estimation to accurately reflect unique impressions, correcting the simple summation approach's inherent inaccuracies.
3Adaptability or versatility
If multiple identifier types are used to define audience segments, then the system can capture more audience characteristics, but the complexity of managing and de-duplicating identifiers increases
Solution Approach 1:
The patent segments identifier management into separate, organized structures for each identifier type (e.g., user IDs, device IDs, IP addresses). Each segment maintains its own data structure and intersection tracking, which simplifies the overall complexity by breaking down the management of multiple identifier types into manageable, parallel operations rather than a single complex process.
4Measurement precision
If advertisers have specific content requirements, then the targeting precision is improved, but the complexity of determining ad placements in finite space increases
Solution Approach 1:
The system segments ad placement decisions by creating separate data structures and processing paths for different advertiser requirements and content types. This allows precise targeting for advertisers with specific content requirements while managing complexity through modular processing of different placement scenarios independently, rather than handling all requirements in a single complex process.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
Systems and methods are provided for efficient computer-implemented forecasting of audience and/or audience demand for supplemental digital content. Specifically, an efficient target segment data structure is used with universe data to identify a viewership supply forecast. The forecast may be used to control downstream forecast dependent systems and / or services.