A BERT based dual-channel explainable text emotion recognition system.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Machine Learning

Background:

• Text emotion recognition is crucial for understanding human-computer interaction.
• Existing models often lack robust explainability for their predictions.
• Multi-class emotion recognition requires sophisticated feature extraction and sequence modeling.

Purpose of the Study:

• To propose a novel dual-channel system for multi-class text emotion recognition.
• To develop a new technique for explaining the system's training and prediction processes.
• To evaluate the system's performance and applicability across diverse datasets.

Main Methods:

• Utilized a pre-trained Bidirectional Encoder Representations from Transformers (BERT) model for feature extraction.
• Employed a dual-channel network combining Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) layers.
• Developed an explainability module analyzing inter- and intra-cluster distances for model transparency.

Main Results:

• Achieved consistent accuracy, precision, recall, and F1 scores across ISEAR, Aman, AffectiveText, and EmotionLines datasets.
• Demonstrated the effectiveness of the dual-channel architecture in capturing textual features and sequential information.
• Validated the proposed explainability technique's ability to interpret model behavior.

Conclusions:

• The proposed dual-channel system offers a robust solution for multi-class text emotion recognition.
• The novel explainability technique provides valuable insights into the model's decision-making process.
• The system's strong performance across multiple datasets confirms its versatility and practical applicability.