Federated multi-label text feature selection via manifold-aware sparse modeling and cooperative grey wolf...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Natural Language Processing

Background:

• Multi-label text classification faces challenges like high dimensionality, label correlation, and feature sparsity.
• Existing feature selection methods are often centralized, unsuitable for distributed or federated environments.
• Lack of effective feature selection mechanisms in decentralized data settings hinders performance.

Purpose of the Study:

• To propose Fed-MSMCGWO, a federated multi-label text feature selection method.
• To address limitations of centralized methods in distributed and federated learning scenarios.
• To enhance feature selection accuracy and preserve data privacy in cross-client settings.

Main Methods:

• Federated learning framework integrating manifold-aware sparse modeling (MSM) and cooperative grey wolf optimization (CGWO).
• Two-stage optimization on each client: Stage 1 uses MSM with graph Laplacians and L1-norm for sparsity; Stage 2 uses CGWO for weight refinement and global search.
• Privacy-preserving feature aggregation strategy where clients upload intermediate weights, not raw data, for server-side aggregation and local updates.

Main Results:

• Fed-MSMCGWO achieves superior performance across multiple evaluation metrics compared to standard centralized and federated feature selection methods.
• The method demonstrates consistent effectiveness on publicly available multi-label text datasets.
• Experimental results confirm the efficacy of the privacy-preserving aggregation strategy.

Conclusions:

• Fed-MSMCGWO offers an effective solution for privacy-preserving multi-label text feature selection in federated environments.
• The integration of MSM and CGWO within a federated framework overcomes challenges of distributed data.
• The proposed method significantly improves feature selection performance while maintaining data confidentiality.