Text Semantic Classification of Long Discourses Based on Neural Networks with Improved Focal Loss

Area of Science:

• Natural Language Processing
• Machine Learning
• Artificial Intelligence

Background:

• Semantic classification of long Chinese discourses presents challenges due to high dimensionality, sparsity, and imbalanced data distributions.
• Existing models struggle to effectively address these issues in complex discourse analysis.

Purpose of the Study:

• To propose a novel end-to-end model, CRAFL, for semantic classification of Chinese long discourses.
• To address data sparsity, high dimensionality, and class imbalance in discourse datasets.

Main Methods:

• Utilized a convolutional layer with an attention mechanism and recurrent neural networks (RNNs).
• Employed Residual Network (ResNet) for phrase semantic representation and dimensionality reduction.
• Incorporated an improved focal loss function to mitigate data class imbalance.

Main Results:

• The CRAFL model demonstrated superior efficiency compared to state-of-the-art models on a long discourse dataset.
• Successfully extracted semantic representations and learned sequence features for improved classification.

Conclusions:

• The proposed CRAFL model offers an effective solution for the semantic classification of Chinese long discourses.
• The integration of attention mechanisms, RNNs, and improved focal loss significantly enhances performance on imbalanced and complex datasets.