Duarte and Berton address the challenges of text classification in the context of big data, emphasizing the difficulties and costs associated with obtaining large labeled datasets. They explore the field of semi-supervised learning (SSL) as a solution, presenting an up-to-date review of its application in text classification. The authors review 1794 works from prominent databases and then select 157 articles for detailed analyses, examining application domains, datasets, languages, text representations, machine learning algorithms, and a recent taxonomy of SSL. The review also discusses the percentage of labeled data used, evaluation metrics, results, limitations, and future trends.

Strengths:

• The paper’s review of 1794 works (that are narrowed down to 157 articles) showcases a thorough and meticulous approach. This comprehensive scope ensures a broad and inclusive understanding of the current state of SSL in text classification.
• By focusing on works from the last five years, the review ensures that it covers the most recent advancements and trends in the field, making it highly relevant for current researchers and practitioners.
• The authors present a well-organized summary of the reviewed works, categorizing them by application domain, datasets, languages, text representations, and machine learning algorithms. This structured approach facilitates a clear and coherent understanding of the various aspects of SSL in text classification.
• The review utilizes a recent taxonomy of SSL, providing a modern framework for organizing and analyzing the works. This enhances the paper’s relevance and applicability to contemporary research.
• By analyzing the percentage of labeled data used, evaluation metrics, and results, the paper provides practical insights into how SSL techniques are implemented and evaluated. This is particularly useful for researchers and practitioners looking to apply SSL in their own work.
• The discussion of current limitations and future trends offers valuable guidance for future research directions. This forward-looking perspective is crucial for advancing the field and addressing existing challenges.

Criticisms:

• While the review is comprehensive, some sections may lack depth in specific areas. For instance, the discussion on text representations and machine learning algorithms could benefit from more detailed explanations and examples to enhance understanding.
• The review mentions the analysis of evaluation metrics but does not provide an in-depth discussion of their effectiveness or limitations. A more critical evaluation of these metrics would be beneficial for understanding the nuances of SSL performance in text classification.
• The paper could further enhance its practical relevance by including more real-world case studies or examples of SSL applications in various industries. This would help bridge the gap between theoretical research and practical implementation.
• While the paper identifies future trends, it could provide more specific predictions or scenarios for how SSL might evolve in the context of text classification. This would offer a clearer roadmap for future research and development.

Duarte and Berton’s paper is a valuable and comprehensive resource for anyone interested in the application of SSL in text classification. The paper excels in its thorough literature review, current relevance, detailed analysis, and practical insights. However, it could benefit from deeper discussions in certain areas, a more critical evaluation of evaluation metrics, additional practical examples, and more specific predictions for future trends.

Overall, the review succeeds in providing a clear and organized overview of the state of SSL in text classification, offering useful information and guidance for researchers and practitioners alike. It is a commendable contribution to the field, addressing a significant gap in the literature and paving the way for future advancements in SSL for text classification.