Dual graph convolutional networks integrating affective knowledge and position information for aspect sentiment...

Area of Science:

• Natural Language Processing (NLP)
• Artificial Intelligence (AI)
• Machine Learning (ML)

Background:

• Aspect Sentiment Triplet Extraction (ASTE) is crucial for analyzing consumer feedback.
• Existing models often overlook affective knowledge and inter-triplet relationships.
• There is a need for advanced NLP models to capture nuanced sentiment information.

Purpose of the Study:

• To propose a novel end-to-end model, DGCNAP, for ASTE.
• To integrate affective knowledge and position information into ASTE models.
• To improve the extraction of aspect-opinion-sentiment triplets from text.

Main Methods:

• Developed Dual Graph Convolutional Networks Integrating Affective Knowledge and Position Information (DGCNAP).
• Incorporated affective knowledge from SenticNet into dependency graph construction.
• Utilized a multi-target position-aware function within the GCN to weigh word proximity.

Main Results:

• DGCNAP significantly outperforms state-of-the-art models on ASTE-Data-V2 datasets.
• Achieved high F1 scores: 70.72 (14res), 57.57 (14lap), 61.19 (15res), and 69.58 (16res).
• Demonstrated the effectiveness of integrating affective and positional information.

Conclusions:

• The proposed DGCNAP model offers a significant advancement in Aspect Sentiment Triplet Extraction.
• Integrating affective knowledge and positional information enhances sentiment analysis accuracy.
• This approach leads to more effective information extraction from comments for improved products and services.