Unsupervised Streaming Feature Selection via Link Information
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional feature selection methods are inadequate for social media, as they assume static features and require labeled data, which is not feasible in high-velocity streaming environments with dynamically generated features and vast amounts of unlabeled data, making unsupervised streaming feature selection challenging due to the lack of label information and non-independent, identically distributed data.
Innovation Solution
The unsupervised streaming feature selection framework (USFS) leverages link information to dynamically select relevant features by extracting social latent factors and using them as constraints in a regression model, integrating link and feature information to efficiently process new and existing features, allowing for timely adaptation to changing data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional feature selection methods are used, then feature selection can be performed with static features, but it cannot adapt to dynamically generated streaming features in social media
Solution Approach 1:
The patent implements dynamic feature selection by processing features in a streaming manner as they arrive, rather than requiring all features to be static and available beforehand. The system continuously updates the selected feature set based on newly arrived features, enabling adaptation to dynamic social media data while maintaining computational feasibility through incremental processing
2Measurement precision
If supervised feature selection is used, then label information can guide feature selection, but it requires time and labor consuming labeled data which is not available in large quantities in social media
Solution Approach 1:
The patent introduces link information as an intermediary substitute for label information. By exploiting the structural relationships and connections between data instances in social media, the system can guide feature selection without requiring explicit labels, thereby maintaining feature selection quality while avoiding the need for large amounts of manually annotated data
3Reliability
If all streaming features are processed, then comprehensive feature selection can be achieved, but processing time increases significantly
Solution Approach 1:
The patent extracts and processes only the most relevant features from the streaming data by applying feature selection criteria to identify and retain important features while discarding redundant ones. This selective extraction approach ensures comprehensive coverage of important features while significantly reducing processing time by avoiding unnecessary computation on all features
4Measurement precision
If feature selection is performed batch-mode, then all features can be evaluated together, but it cannot capture emerging features in real-time
Solution Approach 1:
The patent performs preliminary feature selection by evaluating features as they arrive in the stream and making preliminary decisions about their inclusion in the selected feature set. This preliminary action enables real-time capture of emerging features while maintaining evaluation accuracy through continuous incremental assessment rather than waiting for batch processing
Data Source
AI summary
Systems and methods for exploiting link information in streaming feature selection, resulting in a novel unsupervised streaming feature selection framework are disclosed.