Multi-task learning for aspect level semantic classification combining complex aspect target semantic enhancement and...

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Machine Learning

Background:

• Aspect-based sentiment analysis (ABSA) requires fine-grained sentiment detection.
• Current deep learning models struggle with limited training data for granular ABSA.

Purpose of the Study:

• To develop an enhanced BERT-based model for multi-dimensional aspect target semantic learning.
• To address the scarcity of fine-grained training corpora for ABSA.

Main Methods:

• Leveraging BERT pre-training and fine-tuning for rich semantic features.
• Implementing a complex semantic enhancement for aspect targets to optimize corpora.
• Combining aspect recognition enhancement with a Conditional Random Field (CRF) model for robust entity recognition.
• Utilizing an adaptive local attention mechanism for focused sentiment analysis around aspect targets.
• Optimizing a joint training mechanism for multi-task learning.

Main Results:

• The proposed model significantly improves ABSA performance across multiple Chinese and English datasets.
• The methods demonstrate superior results compared to state-of-the-art models in both multi-task and single-task scenarios.
• Enhanced semantic learning and attention mechanisms contribute to more accurate aspect target identification and sentiment analysis.

Conclusions:

• The enhanced BERT-based model effectively addresses the challenges of fine-grained ABSA.
• Semantic enhancement and adaptive attention mechanisms are crucial for improving ABSA model accuracy.
• The proposed approach offers a robust solution for aspect-based sentiment analysis, particularly with limited granular data.