Reach Forecast Models Using Base Segments and Pre-Trained ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for predicting the efficacy of distribution plans, such as advertising campaigns, are insufficiently accurate and computationally expensive due to the infinite number of variables and lack of training data for each target segment, and the inability to effectively address the need for specific target segments.
Innovation Solution
Use a combination of pre-trained generalized models and data that can be obtained in real-time for specific segments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional forecasting methods are used to predict reach and frequency for specific target segments, then measurement precision is improved, but device complexity and loss of time increase due to the infinite number of variables and lack of training data
Solution Approach 1:
The patent segments the forecasting problem by dividing the infinite space of target segments into a finite set of demographic categories (age groups, gender, geography). Each category is assigned a pre-trained model, transforming an unmanageably complex problem into a structured set of manageable segments that can be processed efficiently
Solution Approach 2:
The patent applies preliminary action by pre-training forecasting models for each demographic category before actual forecasting is needed. This allows the system to have ready-made models for all possible demographic segments, eliminating the need to train models on-demand and significantly reducing both computational complexity and time loss when forecasts are required
2Measurement precision
If traditional forecasting methods are used to predict reach and frequency, then measurement precision is improved, but loss of time increases due to computational expense
Solution Approach 1:
The patent pre-trains forecasting models for all demographic categories in advance, so that when a forecast is needed, the system can immediately apply the appropriate pre-trained model to the input data. This eliminates the time-consuming model training process and allows for rapid forecast generation while maintaining high accuracy
Solution Approach 2:
By segmenting the problem into discrete demographic categories with dedicated pre-trained models, the system can quickly identify and apply the correct model without searching through infinite possibilities, significantly reducing the time required to generate accurate forecasts
3Adaptability or versatility
If traditional forecasting methods are used, then adaptability to specific target segments is improved, but device complexity increases due to the infinite number of variables
Solution Approach 1:
The patent segments the continuous space of target segments into discrete demographic categories (age groups, gender, geography). This segmentation allows the system to adapt to specific target segments by selecting the appropriate pre-trained model for that demographic category, while avoiding the complexity of handling infinite variables through the finite categorical structure
Solution Approach 2:
The patent creates a universal forecasting system where a single framework handles all demographic segments through pre-trained models. Each model is specialized for its demographic category, but collectively they provide universal coverage for any possible target segment, eliminating the need for custom model development for each segment
Data Source
AI summary
Embodiments provide for improved machine learning. A first distribution plan for content is accessed, where the first distribution plan comprises a first target segment and identifies a first set of distribution outlets. A base segment corresponding to the target segment is determined, where the target segment is defined based on a plurality of member attributes and the base segment is defined based on a subset of the plurality of member attributes. A set of forecasts is generated using, for each respective distribution outlet of the first set of distribution outlets, a respective machine learning model trained based on the base segment. A forecasted reach metric for the first distribution is generated plan based on the set of forecasts.