Multi-Model ML Architecture for Media Mix Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing media mix models face challenges in capturing and analyzing diverse data sources, accounting for external factors like economic fluctuations, and adapting to the dynamic nature of digital media channels, which requires processor and memory resource-intensive model updates.
Innovation Solution
The implementation of a processing circuit that obtains interaction data, executes a neural network to generate transformed timeseries emphasis data, and uses a Bayesian regression model to estimate performance variables for media channels, allowing for agile model adaptation and reduced computational costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional media mix models are used to analyze diverse data sources and account for external factors, then measurement precision and model accuracy are improved, but processor and memory resource requirements increase significantly
Solution Approach 1:
The patent divides the media mix model into separate modular components: a neural network module for capturing complex patterns from diverse data sources, a Bayesian regression module for accounting for external factors, and a shape function module for transforming emphasis data. This segmentation allows each component to process specific data types efficiently, reducing overall computational resource requirements while maintaining high measurement precision through specialized processing.
Solution Approach 2:
The patent introduces a shape function as an intermediary component that transforms raw emphasis data into a format suitable for Bayesian regression analysis. This intermediary layer simplifies the relationship between input data and model outputs, enabling more efficient processing while preserving the nuanced relationships between media channel emphasis and performance variables, thus maintaining accuracy with reduced computational cost.
2Adaptability or versatility
If media mix models are updated in real-time to adapt to dynamic media channels, then adaptability is improved, but processor resource consumption increases
Solution Approach 1:
The patent implements dynamic adaptability through the neural network component, which can be trained on historical data and then deployed to process new incoming data in real-time. The model's architecture allows it to adapt to changing media channel characteristics without requiring complete retraining, thus maintaining high adaptability while significantly reducing processor resource consumption during real-time operation.
Solution Approach 2:
The patent performs preliminary training of the neural network and Bayesian regression model on historical data before deployment to production environments. This preliminary action allows the model to capture complex patterns and relationships in advance, enabling real-time processing with reduced computational resources since the heavy lifting of pattern learning has already been completed during the training phase.
3Measurement precision
If complex models are used to capture patterns from diverse data sources, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent segments the complex data analysis task into distinct functional modules: a neural network for pattern recognition from diverse data sources, a shape function for data transformation, and a Bayesian regression for performance prediction. This segmentation reduces overall model complexity by assigning specific responsibilities to each component, making the system more manageable while maintaining high measurement precision through specialized processing at each stage.
Solution Approach 2:
The patent employs parameter changes through the shape function, which transforms emphasis data by adjusting parameters such as saturation levels and response curves. This parameter transformation simplifies the input data structure before it enters the Bayesian regression model, reducing the complexity of the overall model while preserving the nuanced relationships between media channel emphasis and performance variables, thus maintaining accuracy with reduced structural complexity.
Data Source
AI summary
In some aspects, the disclosure is directed to a system. The system can include at least one processing circuit comprising at least one memory and one or more processors, the one or more processors configured to obtain interaction data of a media channel, the interaction data comprising timeseries emphasis data; execute a neural network using the interaction data as input to generate a transformed timeseries emphasis data for the media channel, the neural network configured to estimate a shape function; and execute a Bayesian regression model using the transformed timeseries emphasis data for the media channel as input to generate one or more performance variables for the media channel.