Dynamic Ad Combination Selection via Explore-Exploit Layer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional recommender systems for advertising struggle to maximize revenue by not providing sufficient flexibility in ad display, relying on fixed formats despite having multiple assets for each ad attribute, which limits the potential for maximizing click-through rates and conversion rates.
Innovation Solution
The integration of a thin explore/exploit layer at the frontend ad serving engine allows for dynamic combination distributions of ad assets based on predicted performance, enabling real-time selection of the most effective ad combinations for different user segments and environments, using machine learning to optimize ad rendering.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional recommender systems use fixed ad display formats, then the system complexity is low and implementation is simple, but the adaptability to different user segments and environments is limited, resulting in lower click-through rates and conversion rates
Solution Approach 1:
The patent segments ad assets into multiple attributes (e.g., headline, description, call-to-action button) and creates various combinations of these assets. The explore/exploit layer further segments the selection process into exploration phase (trying new combinations) and exploitation phase (using proven effective combinations), allowing the system to adapt to different user segments while managing complexity through structured organization.
Solution Approach 2:
The patent introduces dynamic ad combination selection where the system can switch between different ad asset combinations based on real-time performance data and user context. The explore/exploit mechanism dynamically adjusts which combinations are tested and which are deployed, enabling the system to adapt to changing user preferences and environmental conditions without requiring complete system redesign.
2Productivity
If multiple ad asset combinations are created and tested, then the click-through rate and conversion rate can be maximized, but the time required for training and the complexity of managing combinations increases
Solution Approach 1:
The patent performs preliminary actions by pre-creating multiple ad asset combinations and pre-training the explore/exploit layer model during off-peak times or in batches. The system pre-processes and organizes combination data, allowing rapid deployment and minimal real-time computation, thus reducing the time loss while maintaining high productivity in terms of click-through and conversion rates.
Solution Approach 2:
The patent ensures continuity of useful action by implementing continuous learning mechanisms where the explore/exploit layer continuously refines its model using feedback from user interactions. The system maintains an ongoing process of A/B testing and performance monitoring, ensuring that ad combination selection remains optimized without requiring periodic restarts or retraining from scratch, thereby reducing time loss while sustaining high productivity.
3Adaptability or versatility
If a thin explore/exploit layer is added to the frontend ad serving engine, then real-time dynamic combination selection is enabled, but the device complexity and processing requirements increase
Solution Approach 1:
The patent extracts the complex combination selection logic into a separate explore/exploit layer that operates independently from the main ad serving engine. This extracted layer handles the computationally intensive tasks of model inference and combination selection, while the main engine focuses on ad retrieval and delivery. By separating these functions, the system achieves real-time dynamic combination selection without proportionally increasing overall processing requirements.
Solution Approach 2:
The patent introduces an intermediary explore/exploit layer that mediates between the ad serving engine and the final ad combination selection. This intermediary layer receives ad assets and user context as input, processes them through the trained model to determine optimal combinations, and outputs the selected combinations for rendering. The intermediary approach allows real-time adaptation while managing processing requirements by pre-computing and caching model predictions where appropriate.
Data Source
AI summary
The present teaching relates to generating combination distributions for ads. Features are computed based on training data associated with ads, each of which has a plurality of attributes. The training data include asset combinations with past performance thereof for each of the ads. Each combination includes multiple assets representing respective attributes of an ad. The features are used in machine learning to obtain an auxiliary model, which is used to generate combination distributions for each ad based on predicted performance for each combination associated with the ad. Such generated combination distributions are sent to an explore/exploit layer (EEL) for a frontend ad serving engine to draw a combination therefrom for an auction winning ad for rendering on a webpage viewed by a user on a user device.