Adversarial Content Revision for Recommendation Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional recommender systems fail to provide users with preferred recommendations as they discard disliked content, leading to imbalanced data classification and inefficient use of training data, especially in scenarios with imperfect labels and imbalanced distributions.
Innovation Solution
The implementation of an adversarial machine learning framework, specifically the Disguise Adversarial Network (DAN), which revises disliked content features to create liked content by transforming major class samples into minor class samples, allowing for improved classification and recommendation accuracy with reduced training data requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional recommender systems discard disliked content, then the system operates with simpler data processing, but recommendation accuracy deteriorates due to imbalanced data classification
Solution Approach 1:
The patent transforms disliked content (harmful factor) into useful training data by applying adversarial revision to modify its features. The content revision system uses neural networks to generate revised versions of disliked content that resemble liked content, converting previously useless data into beneficial training samples that improve recommendation accuracy without requiring additional system complexity
Solution Approach 2:
The system changes the features/parameters of disliked content through adversarial revision. The content revision network modifies content parameters (such as text, images, or audio features) to transform the content from disliked to potentially liked, enabling the system to utilize imbalanced data more effectively while maintaining manageable processing complexity
2Reliability
If conventional recommender systems use imbalanced training data, then data collection is simpler, but classification performance deteriorates
Solution Approach 1:
The patent converts the harmful effect of imbalanced training data into a benefit by using the abundant disliked content as raw material for generating revised training samples. The adversarial revision system transforms disliked content into revised versions that balance the training data distribution, improving classification performance while utilizing the existing imbalanced data quantity
Solution Approach 2:
The system performs preliminary revision of disliked content before it enters the training process. By pre-processing the imbalanced data through adversarial revision to generate balanced training samples, the system prepares the data in advance, ensuring better classification performance without requiring complex real-time balancing mechanisms during training
3Measurement precision
If adversarial revision is applied to transform disliked content, then recommendation accuracy improves, but computational complexity increases
Solution Approach 1:
The patent applies partial adversarial revision by selectively processing only disliked content that needs transformation, rather than revising all content. The content revision network is applied specifically to the minority class (disliked content) to generate balanced training samples, achieving improved recommendation accuracy without the excessive computational cost of processing the entire dataset
Solution Approach 2:
The patent introduces a content revision network as an intermediary component between the data collection stage and the training stage. This intermediary system handles the computationally intensive adversarial revision process separately, allowing the main recommendation system to benefit from improved accuracy while the computational complexity is isolated to the revision module, enabling modular optimization
4Productivity
If disliked content is revised and reused, then data utilization efficiency improves, but processing time increases
Solution Approach 1:
The patent performs adversarial revision as a preliminary action during offline data preparation rather than during online recommendation generation. By pre-revising disliked content into balanced training samples beforehand, the system improves data utilization efficiency for future recommendations while avoiding time delays during actual recommendation serving, as the revision work is completed in advance
Data Source
AI summary
A recommendation method includes retrieving content consumption data including content consumed and content not consumed. Based on the content consumption data, identifying a first piece of content not consumed. A first feature of the first piece of content related to negative consumption of the first piece of content is determined. A first system is used to revise the first feature to a second feature. A second piece of content including the second feature is provided to an electronic device. The second piece of content is a revised instance of the first piece of content.