Improved graph convolutional network for emotion analysis in social media text.

Area of Science:

• Natural Language Processing (NLP)
• Artificial Intelligence (AI)
• Computational Linguistics

Background:

• Understanding emotions in social media text is vital for applications like mental health monitoring and sentiment analysis.
• Existing models often struggle to capture deep semantic relationships within social media data.

Purpose of the Study:

• To develop an Improved Graph Convolutional Network (IGCN) for accurate emotion classification in social media text.
• To enhance the representation of semantic relationships and improve model interpretability.

Main Methods:

• Utilized a Pointwise Mutual Information (PMI) based graph construction method to model word relationships.
• Integrated an attention mechanism to emphasize contextually significant words.
• Applied the IGCN model to large-scale datasets including Twitter_EA and an Emotion Recognition Dataset.

Main Results:

• Achieved classification accuracies of 78.64% and 92.38% on benchmark datasets.
• Demonstrated the effectiveness of graph neural networks (GNNs) for large-scale emotion classification.
• Showcased improved interpretability through attention-weighted word importance.

Conclusions:

• The proposed IGCN model significantly enhances emotion classification accuracy in social media text.
• Graph-based NLP models offer transformative potential for sentiment analysis and understanding emotional tones.
• The model's scalability ensures efficient processing of large social media datasets.