Ad Selection Probability Adjustment via Predicted Performance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Online advertising systems face variability in predicted performance of advertisements across different placements, leading to inconsistent return on investment (ROI), as the relevance of advertisements to user queries can significantly impact their click-through rates and conversion rates.
Innovation Solution
A method to adjust the participation probability of content items, such as advertisements, in a selection process based on predicted performance measures, comparing them to alternative performance measures or thresholds, allowing for continuous scaling and discrete adjustments to optimize their inclusion in auctions and presentations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If advertisements are presented based on aggregate historical performance measures, then the selection process is simple and fast, but the ROI varies significantly due to unpredictable performance in different placements
Solution Approach 1:
The system performs preliminary actions by predicting advertisement performance for each specific placement before the actual selection occurs. The predicted performance measure is calculated in advance for each content item in each selection process, allowing the system to proactively identify and adjust participation probabilities for advertisements that are likely to perform poorly in specific contexts, rather than relying on aggregate historical data alone.
Solution Approach 2:
The system changes the parameter of performance evaluation from static aggregate historical measures to dynamic predicted performance measures that are specific to each placement context. By adjusting the participation probability parameter based on predicted performance comparisons, the system adapts the selection criteria to match the specific characteristics of each selection process, thereby improving ROI consistency.
2Productivity
If all eligible content items are submitted to the selection process, then the system is simple to operate, but low-performing advertisements consume budget and reduce overall campaign effectiveness
Solution Approach 1:
The system applies partial action by selectively adjusting the participation probability of content items rather than uniformly including or excluding all eligible items. By comparing predicted performance measures and adjusting probabilities accordingly, the system applies the necessary level of filtering action - not too little to maintain simplicity, but not too much to sacrifice effectiveness - thereby optimizing campaign productivity while maintaining reasonable operational simplicity.
Solution Approach 2:
The system implements feedback by continuously comparing predicted performance measures against alternative measures and using this information to adjust participation probabilities. This feedback mechanism allows the system to automatically identify and limit the selection of low-performing advertisements, improving campaign effectiveness without requiring complex manual intervention, as the adjustment process is driven by automated performance comparisons.
3Reliability
If participation probability is adjusted based on predicted performance comparisons, then low-performing ads are limited, but the system requires complex performance measurement and comparison mechanisms
Solution Approach 1:
The system performs preliminary performance prediction for each content item in each selection process before the actual selection occurs. By calculating predicted performance measures in advance and comparing them to alternative measures, the system proactively identifies advertisements that are likely to perform poorly, allowing for early adjustment of participation probabilities and avoiding the need for complex real-time measurement and adjustment mechanisms.
Data Source
AI summary
Methods, systems, and apparatus, including computer program products, in which participation probabilities for content items in content item selection processes are adjusted based upon predicted performance measures. The predicted performance measures can be compared to one or more other predicted performance measures and/or threshold/scaling data to determine how to adjust the participation probability of the content item in the content item selection process.