Chinese text classification method based on sentence information enhancement and feature fusion

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Artificial Intelligence

Background:

• Chinese text classification faces challenges due to complex semantics and feature extraction difficulties.
• Traditional methods struggle with word-sentence relationships, limiting deep semantic understanding and performance.
• Existing models often fail to effectively filter irrelevant information in Chinese text.

Purpose of the Study:

• To propose a novel Chinese text classification method addressing semantic and feature extraction challenges.
• To enhance the representation of Chinese text by incorporating utterance information and fusing diverse features.
• To improve the accuracy and effectiveness of Chinese text classification models.

Main Methods:

• Utilized BERT (Bidirectional Encoder Representations from Transformers) for text embedding and initial feature extraction (word and sentence vectors).
• Developed an utterance information enhancement module for syntactic enhancement and sentence-level feature extraction.
• Implemented a feature fusion strategy combining enhanced sentence features with Bi-GRU (Bidirectional Gated Recurrent Unit network) word-level features.

Main Results:

• The proposed method demonstrated superior performance compared to existing mainstream classification models.
• Achieved higher classification accuracy and F1 values on multiple Chinese datasets.
• Effectively enhanced feature representation and filtered out irrelevant information in Chinese text.

Conclusions:

• The proposed utterance information enhancement and feature fusion method is effective and feasible for Chinese text classification.
• This approach significantly improves the ability to capture deep semantic information in Chinese text.
• The method offers a promising solution for overcoming the limitations of traditional Chinese text classification techniques.