Joint local and global consistency on interdocument and interword relationships for co-clustering

Area of Science:

• Data Mining
• Machine Learning
• Natural Language Processing

Background:

• Co-clustering partitions words and documents simultaneously, leveraging their relationships.
• Existing methods often overlook or incompletely capture interword and interdocument dependencies.

Purpose of the Study:

• To propose a novel co-clustering method that fully exploits word-document relationships.
• To integrate local and global consistencies within both word and document spaces.

Main Methods:

• Word-document relationships modeled using Information-Theoretic Co-Clustering (ITCC).
• Local consistency captured via a local predictor for interword and interdocument relationships.
• Global consistency enforced through regularization on the data manifold.

Main Results:

• An integrated objective function combining fitting errors from ITCC, local, and global consistency.
• Iterative optimization with a provably convergent updating procedure.
• Validation of effectiveness on benchmark document datasets.

Conclusions:

• The proposed method effectively models word-document relationships.
• Jointly exploring local and global consistencies improves co-clustering performance.
• Demonstrated effectiveness on real-world document datasets.