Automated Persona Feature Selection for Real-Time Content Customization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing content customization methods in real-time environments face challenges in rapidly selecting relevant features for soft clusters due to resource-intensive behavioral models and low-frequency features, which can lead to delayed content delivery and inefficient targeting.
Innovation Solution
A method that generates compact feature sets using a Bernoulli mixture model with nearest shrunken centroids to select prevalent and discriminating features, enabling rapid similarity assessment and personalized content delivery without interfering with the user's browsing experience.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large behavioral model is used to improve cluster selection accuracy, then the applicability to a larger audience is improved, but the time and resources required to operate the model increase, making it unrealistic for real-time applications
Solution Approach 1:
The patent extracts and selects only the most relevant features from the complete feature set to create a reduced feature set. This extraction process identifies features with high discriminatory power for cluster selection while removing redundant or low-value features, thereby reducing model complexity and operation time while maintaining accuracy
Solution Approach 2:
The patent changes the parameter of feature dimensionality by transforming the complete feature set into a reduced feature set through statistical analysis. By modifying the number and selection of features used in the model, the system achieves faster operation times while preserving the essential information needed for accurate cluster selection
2Measurement precision
If multiple large behavioral models are applied to improve targeting accuracy, then the precision of audience segmentation is improved, but the computational resources and time required become unrealistic for real-time environments
Solution Approach 1:
The patent segments the complete feature set into distinct groups based on their relevance and discriminatory power for different clusters. By dividing the features into retained and excluded groups, the system creates a streamlined model that maintains segmentation precision while reducing computational resource requirements
Solution Approach 2:
The patent creates a simplified copy of the behavioral model using only the most essential features. This reduced-feature model replicates the core functionality of the complete model for cluster selection purposes, consuming fewer computational resources while achieving comparable segmentation accuracy
3Productivity
If a compact feature set is used to reduce model operation time, then the real-time performance is improved, but the applicability to diverse audiences may be reduced
Solution Approach 1:
The patent carefully adjusts the parameter of feature selection criteria to retain only those features that demonstrate both high discriminatory power and broad applicability across diverse audiences. By changing the selection thresholds and criteria, the system achieves a compact feature set that maintains real-time performance while preserving audience coverage
Data Source
AI summary
Embodiments of the invention include a system for automated persona feature selection. Soft clusters of entities are received, each entity having a history of features. Each feature has a general prevalence coefficient representing prevalence of entities having the respective feature in their history. A feature list is generated for each cluster, each feature having an in-cluster coefficient representing prevalence of entities in the cluster having the feature in their history. Features having an in-cluster coefficient that is different from that feature's general prevalence coefficient are selected. A variance across the clusters is determined for each selected feature. A discriminating feature list having high variance features is generated for each cluster. Clusters are selected for an entity by comparing the features of the entity's history to features of the discriminating feature lists of the clusters. Content is customized according to the chosen clusters and sent to the entity.