Sequential Multi-Task Ranking With Category-Aware Feature Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning algorithms for ranking listings on exchange platforms fail to effectively consider the sequential nature of multiple tasks and category-based variations, leading to inaccuracies in ranking due to uniform and non-uniform feature distributions across regions.
Innovation Solution
Implement machine learning models that treat tasks as sequential tasks, separate input features into category-invariant and category-variant features, and use category embeddings to transform category-variant features, combining them with category-invariant features in a trained neural network for accurate scoring.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning algorithms are used for ranking listings, then the system can process multiple tasks simultaneously, but the accuracy of ranking scores deteriorates due to failure to capture sequential relationships between tasks and category-based variations
Solution Approach 1:
The patent segments the input features into category-invariant features and category-variant features. Category-invariant features are processed by a shared neural network, while category-variant features are processed by separate category-specific neural networks. This segmentation allows the system to capture sequential relationships and category-based variations without requiring a single overly complex model, thereby improving ranking accuracy while managing model complexity.
2Productivity
If traditional machine learning approaches are used, then the computation process is simpler, but the computation speed deteriorates due to redundant computations across multiple tasks
Solution Approach 1:
The patent implements preliminary action by pre-processing and separating features into category-invariant and category-variant groups before entering the neural network. This allows the shared neural network to process common features once for all tasks, and only the category-specific networks need to process their respective variant features. This preliminary organization eliminates redundant computations that would occur in traditional approaches where all tasks process all features independently, thereby increasing computation speed.
3Adaptability or versatility
If the model processes all features uniformly for all tasks, then the implementation is simpler, but the ability to handle category-based variations deteriorates
Solution Approach 1:
The patent applies local quality by making different parts of the feature processing system have different properties. Category-invariant features are processed uniformly by a shared neural network with uniform properties, while category-variant features are processed by category-specific neural networks with properties tailored to each category's characteristics. This allows the system to adapt to category-based variations in feature distributions while maintaining simplicity in the uniform processing path, thereby improving adaptability without excessive complexity.
Data Source
AI summary
Methods, system, and apparatus for providing receiving, from a user device, at a computing server, and during a web session on an exchange platform, a set of data including data related to a user-submitted query and interaction data. Using the set of data, a first set of listings is identified that are responsive to the user-submitted query and that correspond listings of a digital component on the platform. The server inputs to a neural network (NN) a set of sequential input features, where the neural network is trained to perform k tasks based on a common set of input data, wherein the k tasks include (1) predicting interaction activity for a candidate listing and (2) predicting affirmative action activity for the candidate listing. The NN generates a set of sequential output scores, which can be used to generate a ranked set of listings.