A survey on text classification: Practical perspectives on the Italian language.

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Machine Learning

Background:

• Deep learning has revolutionized text classification, primarily in English.
• Limited resources and linguistic complexities hinder adoption in other languages.
• Italian language text classification lacks extensive research and benchmarked datasets.

Purpose of the Study:

• To survey challenges in applying modern text classification to non-English languages, focusing on Italian.
• To highlight issues of dataset scarcity and computational expense.
• To propose a linguistically inclusive approach to text classification.

Main Methods:

• Comparative analysis of dataset availability for Italian and French.
• Application of representative text classification methods to custom multilabel datasets.
• Datasets created in Italian, French, and English for practical scenario simulation.

Main Results:

• Identified significant challenges in Italian text classification, including data scarcity.
• Demonstrated the impact of linguistic variations on model performance.
• Compared computational costs of modern approaches across languages.

Conclusions:

• Addressing data scarcity and computational challenges is crucial for non-English text classification.
• Further research is needed for linguistically inclusive and equitable NLP development.
• Future work should focus on creating diverse datasets and efficient models for under-resourced languages.