Conversion Model Training with Outlier Bias Calibration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing content serving systems use conversion probabilities based on data with unforeseen bias, leading to inaccurate bid values due to inclusion of outlier conversion events that lack affinity with clicks, resulting in less accurate user intention reflection.
Innovation Solution
A system that filters out outlier conversion events during model training and incorporates them as a bias for predicting conversion rates, using a machine learning model trained on filtered data to select content items for presentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If all conversion events are used for model training, then more data is available for training, but the accuracy of bid values deteriorates due to inclusion of outlier events with unforeseen bias
Solution Approach 1:
The patent extracts outlier conversion events from the complete dataset using statistical methods (e.g., identifying events where the conversion time deviates significantly from the expected distribution). These outliers are separated into a distinct subset, allowing the main model training to proceed with only typical conversion events, thereby eliminating the bias they introduce while preserving the majority of useful training data.
Solution Approach 2:
The patent segments the conversion event data into two distinct subsets: typical conversion events and outlier conversion events. This segmentation allows different processing approaches for each subset - typical events are used for model training while outliers are handled separately through bias calculation, resolving the contradiction between data quantity and accuracy.
2Quantity of substance
If outlier conversion events are included in bid value calculations, then more comprehensive data is used, but user intention reflection deteriorates due to lack of affinity between outlier events and clicks
Solution Approach 1:
The patent converts the harmful effect of outlier events into a beneficial correction mechanism. Instead of discarding outlier events entirely, the system calculates a bias value from these outliers and applies it as a correction to the model's predictions. This transforms the previously harmful biased data into a useful calibration tool that improves overall accuracy.
Data Source
AI summary
In an example, sets of event information associated with events may be identified. The events may include conversion events having affinity between conversion actions and associated ad clicks and outlier conversion events. Outlier bias associated with the outlier conversion events may be determined for content items. Machine learning model training may be performed, using only the sets of event information associated with the conversion events having affinity between conversion actions and associated ad clicks, to generate a machine learning model. Conversion probabilities associated with content items may be determined using the machine learning model and the calculated biases. A content item may be selected for presentation via a client device based upon the conversion probabilities.