Research on sentiment classification for netizens based on the BERT-BiLSTM-TextCNN model.

Area of Science:

• Computational Linguistics
• Natural Language Processing
• Artificial Intelligence

Background:

• Online public opinion sentiment analysis is crucial for risk management.
• Existing methods struggle with short, unstructured, and syntactically varied comments.
• Challenges include open word ranges and reversed word order in user-generated text.

Purpose of the Study:

• To propose an effective hybrid model for sentiment classification of online comments.
• To address limitations of current methods in handling short and complex text.
• To improve the accuracy and stability of multi-class emotion classification.

Main Methods:

• Developed a hybrid sentiment classification model: BERT-BiLSTM-TextCNN.
• Utilized Bidirectional Encoder Representations from Transformers (BERT) for word vector generation, capturing contextual semantics.
• Integrated Bidirectional Long Short-Term Memory (BiLSTM) for contextual information and Text Convolutional Neural Network (TextCNN) for feature extraction.

Main Results:

• The BERT-BiLSTM-TextCNN model demonstrated superior accuracy and stability compared to existing methods.
• The hybrid model effectively combines local correlations with context information.
• Achieved more flexible extraction of emotional information for accurate multi-class emotion classification.

Conclusions:

• The proposed BERT-BiLSTM-TextCNN model significantly improves multi-label emotion classification accuracy.
• BERT provides rich prior information and contextual semantics.
• BiLSTM and TextCNN modules enhance contextual understanding and feature extraction, respectively.