Separate Conversion Models for Bidding Differentiation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing online advertising campaign models fail to provide finer grain differentiation between bidding opportunities due to features like previous website visits overshadowing other signals, leading to suboptimal conversion prediction capabilities.
Innovation Solution
Building separate models to predict conversion likelihood based on the presence or absence of specific features, such as previous visits to an advertiser's website, allowing for tailored scoring of entity consumption histories to determine conversion probabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single model is formed for each advertising campaign to assess bidding opportunities, then the model covers all possible scenarios, but certain highly relevant features (such as previous website visits) overshadow other signals and reduce differentiation capability
Solution Approach 1:
The patent divides a single advertising campaign model into multiple specialized models based on entity characteristics (e.g., converters vs. non-converters, website visitors vs. non-visitors). Each model is trained on specific subsets of data with particular feature profiles, allowing them to capture nuanced patterns that would be diluted in a unified model. This segmentation enables finer grain differentiation between bidding opportunities while maintaining comprehensive coverage through model selection based on entity features.
2Measurement precision
If separate models are built for different entity types (e.g., converters vs. non-converters), then finer grain differentiation is achieved, but the system complexity increases
Solution Approach 1:
The patent performs preliminary classification of entities into distinct groups (converters, non-converters, website visitors, non-visitors) before model application. Entity features are pre-processed and stored in a structured format that enables rapid model selection. This preliminary organization of data and entity characteristics simplifies the runtime decision process, reducing the operational complexity of managing multiple models despite the increased model diversity.
Data Source
AI summary
Separate models are built to predict the likelihood of conversion based on the presence or absence of one or more features. For example, a first model may be built to predict the likelihood of conversion of a non-converter who has never visited an advertiser's website before and a second model may be built to predict the likelihood of conversion of a non-converter who has visited an advertiser's website before. To determine which model to apply to an entity, the consumption history of the entity is searched for the presence or absence of the one or more features used to separate the models. The entity's consumption history is then scored based on the applicable model to determine the likelihood of conversion.