A multi-label text sentiment analysis model based on sentiment correlation modeling.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Machine Learning

Background:

• Sentiment analysis is crucial for understanding text, but traditional models struggle with multi-label classification and nuanced emotion correlations.
• Existing models often lack interpretability and struggle with the complexities of diverse emotional expressions within text.

Purpose of the Study:

• To introduce the Emotion Correlation-enhanced Sentiment Analysis Model (ECO-SAM) for improved multi-label sentiment analysis.
• To enhance the modeling of semantic correlations between emotions within text data.
• To achieve interpretable sentiment analysis with superior classification performance.

Main Methods:

• Utilized a pre-trained BERT encoder for semantic text embedding.
• Employed a self-attention mechanism to model semantic correlations between emotions.
• Developed a text emotion matching neural network for sentiment classification.

Main Results:

• ECO-SAM demonstrated significant improvements over baseline models on public datasets.
• Achieved up to a 13.33% increase in precision, 3.69% in recall, and 8.44% in F1 score.
• The model's sentiment semantics were found to be interpretable.

Conclusions:

• ECO-SAM effectively models sentiment semantics and achieves excellent classification performance.
• The model shows promise for various public sentiment analysis applications.
• Future work could explore multi-modal data and larger training datasets.