Assemble the shallow or integrate a deep? Toward a lightweight solution for glyph-aware Chinese text classification

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Machine Learning

Background:

• Hieroglyphic languages like Chinese possess unique glyph features that can enhance semantic representation.
• Existing models for glyph-aware text classification are often computationally expensive, limiting their practical application.
• There is a need for efficient methods that balance performance and computational cost in glyph-aware Chinese text classification.

Purpose of the Study:

• To develop a lightweight and computationally efficient method for glyph-aware Chinese text classification.
• To demonstrate that ensemble learning with shallow networks can achieve competitive results against large-scale models.
• To highlight the significance of glyph features in representing hieroglyphic languages.

Main Methods:

• Proposed a lightweight ensemble learning method for glyph-aware Chinese text classification (LEGACT).
• Utilized typical shallow neural networks as base learners.
• Employed machine learning classifiers as meta-learners for the ensemble.

Main Results:

• The LEGACT method achieved comparable performance to large-scale transformer models in glyph-aware Chinese text classification.
• Demonstrated the effectiveness of integrating shallow neural networks through an ensemble approach.
• Provided empirical evidence for the importance of glyph features in hieroglyphic language representation.

Conclusions:

• The proposed LEGACT method offers a lightweight yet powerful solution for glyph-aware Chinese text classification.
• Ensemble strategies combining shallow neural networks are effective in reducing computational workload for predictive tasks.
• Glyph features are crucial for robust semantic representation in hieroglyphic languages.