Flexible Multi-Task Neural Network for Content Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network models for content ranking in online systems face high computational requirements, limiting their ability to develop large neural network structures and train models with large volumes of data, which restricts their predictive capabilities for user interactions with content items.
Innovation Solution
A flexible multi-task neural network prediction model is employed, comprising independent layers for task-specific feature extraction, shared layers for common feature extraction across tasks, and separate layers for predicting user interactions, allowing for efficient ranking of content items based on predicted likelihoods of user actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a neural network model with large structure is used to predict user interactions, then predictive accuracy improves, but computational requirements and processing unit usage increase
Solution Approach 1:
The patent divides the neural network into multiple independent modules, each responsible for specific feature extraction tasks. This segmentation allows the system to process different aspects of user interaction predictions separately, reducing the computational burden on any single module while maintaining overall predictive accuracy through the aggregation of results from multiple specialized modules.
2Measurement precision
If a neural network model with large structure is developed, then predictive capabilities improve, but device complexity increases
Solution Approach 1:
The patent designs independent modules that can perform multiple functions within the neural network system. Each module is capable of extracting different types of features and can be applied to various prediction tasks, reducing the need for separate specialized modules for each function. This multi-functionality simplifies the overall system architecture while maintaining comprehensive predictive capabilities.
3Measurement precision
If large volume of training data is used to train neural network, then model performance improves, but training time and computational resources increase
Solution Approach 1:
The training process is divided into multiple independent training stages, with each stage focusing on training specific modules with relevant subsets of training data. This segmented approach allows parallel processing of different module training, reducing overall training time while ensuring each module learns from the most pertinent data, thereby maintaining model performance.
Data Source
AI summary
For a content item with unknown tasks performed by a viewing user on an online system, the online system predicts a likelihood of interacting with each content item using a prediction model associated with a plurality of tasks. The prediction model comprises a plurality of independent layers, a plurality of shared layers and a plurality of separate layers. Each independent layer is configured to extract features, for each task, that are not shared across the plurality of tasks. The plurality of shared layers are configured to extract common features that are shared across the plurality of tasks. Each separate layer is configured to predict likelihood of the viewing user performing a task associated with the separate layer based on the features extracted from the plurality of independent layers and the plurality of shared layers.