Concept Factorization With Adaptive Neighbors for Document Clustering.

Area of Science:

• Machine Learning
• Data Mining
• Information Retrieval

Background:

• Concept Factorization (CF) is crucial for dimensionality reduction and feature extraction.
• Existing graph-regularized CF methods struggle to simultaneously preserve data geometry and perform dimensionality reduction.
• There is a need for advanced CF techniques that maintain the intrinsic neighborhood structure of data.

Purpose of the Study:

• To propose a novel concept factorization method, CF with adaptive neighbors (CFANs).
• To integrate adaptive neighbors (ANs) regularization into CF decomposition to extract a representation space preserving geometrical neighborhood structure.
• To develop an efficient algorithm for solving the proposed CFAN problem.

Main Methods:

• CFANs integrates an ANs regularization constraint into the CF decomposition process.
• It simultaneously performs dimensionality reduction and learns a neighbor graph weights matrix.
• An efficient algorithm is derived to solve the optimization problem.

Main Results:

• CFANs successfully extracts a representation space that maintains the geometrical neighborhood structure of the data.
• The method was applied to document clustering on 20 Newsgroups, Reuters-21578, and TDT2 datasets.
• Experimental results demonstrate the effectiveness of CFANs in document clustering tasks.

Conclusions:

• CFANs offers a novel and effective approach to concept factorization by preserving data geometry.
• The simultaneous dimensionality reduction and neighbor graph learning is a key advancement.
• The method shows significant promise for applications like document clustering.