BHGAttN: A Feature-Enhanced Hierarchical Graph Attention Network for Sentiment Analysis.

Area of Science:

• Natural Language Processing
• Deep Learning
• Artificial Intelligence

Background:

• Deep learning has accelerated text classification.
• Existing methods often overlook text's hierarchical structure and inter-sentence connections.
• A need exists for models that leverage textual hierarchy.

Purpose of the Study:

• To propose a Bert-based hierarchical graph attention network (BHGAttN) model.
• To effectively model the hierarchical relationships within texts.
• To improve text classification performance by considering sentence-level dependencies.

Main Methods:

• Utilized a large-scale pre-trained BERT model.
• Incorporated a graph attention network to model hierarchical text structures.
• Enhanced semantic features using BERT's intermediate layer outputs.
• Constructed multilevel hierarchical graph networks based on sentence dependencies.

Main Results:

• The BHGAttN model demonstrated significant competitive advantages over state-of-the-art baseline models.
• Experimental results validate the model's effectiveness in text classification.
• The model successfully captures layer-by-layer semantic information and hierarchical relationships.

Conclusions:

• The proposed BHGAttN model offers a superior approach to text classification.
• Modeling hierarchical text structures enhances classification performance.
• BHGAttN represents a significant advancement in deep learning for text analysis.