Thematic clustering of text documents using an EM-based approach

Area of Science:

• Computer Science
• Information Retrieval
• Natural Language Processing

Background:

• Text clustering is crucial for information mining but often lacks human interpretability.
• Existing methods struggle to explain cluster subjects without semantic resources like ontologies.
• A need exists for text clustering that balances performance with explainability.

Purpose of the Study:

• To present a novel thematic clustering algorithm for text documents.
• To enable human understanding of cluster subjects.
• To achieve efficient and competitive clustering performance.

Main Methods:

• Subject terms are extracted from text documents.
• A probabilistic framework and Expectation-Maximization (EM) algorithm are employed for clustering.
• Documents are assigned to clusters based on extracted subject terms.

Main Results:

• The algorithm generates explanatory results for human comprehension.
• The method demonstrates efficient clustering performance.
• Experiments on the MEDLINE dataset show competitive results against state-of-the-art approaches.

Conclusions:

• The proposed thematic clustering algorithm offers a human-interpretable and efficient solution.
• Extracted themes effectively represent cluster subjects, as shown on the MEDLINE dataset.
• This approach advances text mining by improving the explainability of clustering results.