Ad Ranking Using Ghost Content Items for Causal Measurement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining the effectiveness of content items in a computer network environment face challenges such as noise and bias due to random population selection and non-random assignment to test or control groups, leading to inaccurate measurements of conversions and inefficient advertising strategies.
Innovation Solution
A system and method that utilize a data processing system to predict a 'ghost' content item likely to win an auction, allowing for the selection of client computing devices for measurement based on predicted exposure, thereby reducing noise and bias by ensuring proper group assignment and only charging for the ghost content item, and using a cost-per-incremental-action bidding model to align advertiser, publisher, and viewer incentives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If random population selection and non-random assignment to test or control groups are used, then the measurement process is simple, but the measurement precision deteriorates due to noise and bias
Solution Approach 1:
The system performs preliminary actions by predicting auction outcomes and determining counterfactual exposure status before conducting the actual measurement. This allows for proper group assignment that eliminates selection bias and ensures accurate measurement of incremental conversions.
Solution Approach 2:
The system introduces an intermediary mechanism (the measurement system with ghost content items) that mediates between the advertisement auction process and the conversion measurement process. This intermediary ensures proper experimental design by controlling group assignment based on predicted exposure status.
2Adaptability or versatility
If traditional bidding models are used, then the auction process is simple, but the alignment of incentives between advertisers, publishers, and viewers deteriorates
Solution Approach 1:
The system changes the bidding parameter from traditional metrics to cost-per-incremental-action (CPA) bidding based on predicted incremental conversions. This parameter change aligns incentives by making advertisers pay for actual incremental value rather than impressions or clicks, while the measurement system provides the necessary data to support this bidding model.
3Measurement precision
If ghost content items are predicted and charged for, then the accuracy of effectiveness measurement improves, but the computational complexity increases
Solution Approach 1:
The system extracts only the necessary information for measurement accuracy by predicting only the ghost content item that would have won the auction, rather than simulating entire auction processes. This extraction approach reduces computational complexity while maintaining measurement precision by focusing on the critical counterfactual element.
Data Source
AI summary
Systems and methods of serving advertisements in a computer network environment are provided. A data processing system can receive a request for content from a client computing device. The data processing system can receive a cost-per-incremental-action (“CPIA”) bid from a content provider computing device. The data processing system can calculate a value of expected incremental actions from serving a first candidate online ad corresponding to the CPIA bid, and use it to translate the CPIA bid to an auction bid. The data processing system can select a winning online ad from among the first candidate online ad and additional candidate online ads based on their respective corresponding auction bids. The data processing system can transmit the winning online ad to the client computing device.