Nonnegative Matrix Factorization with Rank Regularization and Hard Constraint.

Area of Science:

• Data Science
• Machine Learning
• Dimensionality Reduction

Background:

• Nonnegative matrix factorization (NMF) is a key technique for dimensionality reduction in big data applications.
• Existing NMF methods struggle with optimal prior information utilization, sparse constraint impacts on manifold structure, and graph construction accuracy.

Purpose of the Study:

• To introduce a novel semisupervised NMF algorithm, NMFRC, addressing limitations of traditional NMF.
• To enhance data representation, interpretability, and clustering accuracy in big data analysis.

Main Methods:

• Developed nonnegative matrix factorization with rank regularization and hard constraint (NMFRC).
• Incorporated label information as a hard constraint for improved prior information usage.
• Utilized geodesic distance for pairwise similarity measurement, enhancing manifold information.
• Employed rank constraint for regularization to balance data sparseness and smoothness.

Main Results:

• NMFRC demonstrated superior performance compared to four state-of-the-art algorithms.
• The algorithm achieved higher clustering accuracy on real-world datasets.
• NMFRC provides a more representative and interpretable data representation.

Conclusions:

• NMFRC offers a significant advancement in semisupervised NMF for big data.
• The proposed method effectively overcomes limitations of existing NMF algorithms.
• NMFRC shows strong potential for applications requiring accurate data clustering and analysis.