Shared Prediction Models for Content Clusters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Constructing a single machine-learned model for multiple categories in online systems, such as social networking or advertising, often results in reduced prediction accuracy due to disregarding category differences, while building separate models for each category requires significant computational resources.
Innovation Solution
The online system generates shared models for clusters of categories, where each shared model is specific to the group of categories within a cluster, providing good prediction accuracy while minimizing computational complexity, and also maintains a general model for cases where shared models are not applicable or when a category is unassigned.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by stationary object
If a single machine-learned model is constructed for multiple categories, then computational resources required for model maintenance are reduced, but prediction accuracy deteriorates due to disregarding differences between individual categories
Solution Approach 1:
The patent segments the single model into multiple shared models, each serving a cluster of categories. Categories are grouped into clusters based on their characteristics, and each cluster has its own shared model trained on data from all categories in that cluster. This segmentation allows the system to capture category-specific patterns while sharing data across similar categories, thereby improving prediction accuracy without requiring separate models for every category.
Solution Approach 2:
The patent applies local quality by making each shared model specific to its cluster of categories rather than using a single uniform model for all categories. Each shared model is trained on data from its specific cluster, allowing it to adapt to the local characteristics and patterns of that particular group of categories, thus improving prediction accuracy for category-specific scenarios while still maintaining efficiency.
2Measurement precision
If separate machine-learned models are constructed for each category, then prediction accuracy is improved by accounting for category differences, but computational complexity and resource requirements increase significantly
Solution Approach 1:
The patent merges multiple category-specific models into shared models that serve clusters of categories. Instead of maintaining separate models for each category, the system combines data from multiple categories within a cluster to train a single shared model. This merging reduces the total number of models from N (number of categories) to K (number of clusters, where K < N), significantly reducing computational complexity while still accounting for category differences through cluster-specific training.
Solution Approach 2:
The patent makes each shared model universal by designing it to serve multiple categories within its cluster. Each shared model is trained on aggregated data from all categories in its cluster, enabling it to function as a multi-functional model that can predict outcomes for any category within that cluster. This universality reduces the overall number of models required while maintaining the ability to handle category-specific variations.
3Reliability
If category-specific models are used, then differences between categories are captured improving accuracy, but the system requires significant computational resources to maintain multiple models
Solution Approach 1:
The patent segments the model maintenance burden by organizing categories into clusters and assigning shared models to each cluster. This segmentation reduces the total number of models from N individual category models to K cluster-based shared models (where K < N), directly reducing computational resources required for model training, validation, and deployment while still capturing category-specific patterns through cluster-oriented training.
Solution Approach 2:
The patent merges the training data and model maintenance responsibilities across multiple categories within each cluster. By combining data from all categories in a cluster to train a single shared model, the system reduces redundant computations that would occur if separate models were maintained for each category. This merging approach maintains reliability by preserving category differences through cluster-specific training while reducing the overall computational burden.
Data Source
AI summary
An online system, such as a social networking system, generates shared models for one or more clusters of categories. A shared model for a cluster is common to the categories assigned to the cluster. In this manner, the shared models are specific to the group of categories (e.g., selected content providers) in each cluster while requiring a reasonable computational complexity for the online system. The categories are clustered based on the performance of a model specific to a category on data for other categories.