Long text feature extraction network with data augmentation.

Area of Science:

• Natural Language Processing
• Information Science
• Public Health

Background:

• The COVID-19 pandemic increased reliance on social media for information, leading to widespread misinformation.
• Existing Chinese datasets are insufficient for detecting epidemic-related fake news due to data limitations.
• Long text data poses challenges in classification due to the loss of edge characteristics.

Purpose of the Study:

• To propose a novel network for extracting features from long Chinese text data for fake news detection.
• To improve the accuracy and learning performance of fake news classification models.
• To address the challenges of data imbalance and noise in Chinese epidemic datasets.

Main Methods:

• A long text feature extraction network with data augmentation (LTFE) was developed.
• Twice-Masked Language Modeling for Fine-tuning (TMLM-F) and Data Alignment that Preserves Edge Characteristics (DA-PEC) were employed for feature extraction.
• Attention mechanisms were used to capture word dependencies and generate vector representations.

Main Results:

• The proposed LTFE method demonstrated effectiveness in detecting Chinese fake news related to the pandemic.
• The approach improved the learning performance of the classifier by optimizing data feature structures.
• The combination of TMLM-F and DA-PEC successfully extracted relevant classification features.

Conclusions:

• The developed LTFE method is effective for identifying fake news in Chinese social media during the COVID-19 pandemic.
• The proposed techniques enhance the classification of long text data by preserving crucial edge characteristics.
• This research offers a valuable tool for combating misinformation in public health crises.