Joint coordinate attention mechanism and instance normalization for COVID online comments text classification.

Area of Science:

• Natural Language Processing
• Machine Learning

Background:

• Traditional text classification methods often focus on distinct features, leading to computational inefficiencies.
• A novel approach is proposed to leverage label information directly for constructing text representations.
• This aims to optimize the use of label data alongside textual content for improved classification.

Purpose of the Study:

• To develop a computationally efficient text classification methodology.
• To enhance classification accuracy by integrating label information into text representations.
• To address the limitations of existing feature extraction-focused approaches.

Main Methods:

• Separate pre-processing of texts and labels, followed by encoding via a projection layer.
• Utilized a self-attention model enhanced with instance normalization (IN) and Gaussian Error Linear Unit (GELU) for emotional valence assessment.
• Developed an advanced self-attention mechanism for efficient text-label integration and employed an adaptive label encoder.

Main Results:

• The proposed model demonstrated significant improvements in text classification performance.
• Achieved superior micro-F1 scores compared to existing methodologies.
• Empirical evaluations confirmed the model's efficacy in addressing computational inefficiencies and enhancing accuracy.

Conclusions:

• Integrating label information directly into text representations is effective for improving classification performance.
• The novel approach offers a more efficient and accurate solution for text classification tasks.
• The study highlights the potential of label-aware text representation learning.