Word distance assisted dual graph convolutional networks for accurate and fast aspect-level sentiment analysis

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Machine Learning

Background:

• Aspect-level sentiment analysis offers granular sentiment classification for specific topics.
• Graph Convolutional Networks (GCNs) are popular for capturing word correlations in sentiment analysis.
• Existing GCN methods often overlook word distance, leading to misclassifications.

Purpose of the Study:

• To propose a novel dual GCN structure that integrates word distance, syntactic information, and sentiment knowledge.
• To address the cross-misclassification issue caused by ignoring word distance in GCNs.
• To enhance the accuracy and efficiency of aspect-level sentiment analysis.

Main Methods:

• Developed a dual GCN structure incorporating word distance.
• Utilized word distance to enhance syntactic dependency trees.
• Constructed a new graph with semantic knowledge using word distance.
• Fed two word distance-assisted graphs into separate GCNs for classification.

Main Results:

• Achieved higher classification accuracy compared to state-of-the-art methods.
• Demonstrated significant training acceleration, up to 1.81x.
• Validated the approach on both Chinese and English datasets, including MOOC comments and Douban book reviews.

Conclusions:

• The proposed dual GCN structure effectively leverages word distance, syntactic, and semantic information for improved aspect-level sentiment analysis.
• The method offers a promising solution for accurate and efficient sentiment classification.
• The approach shows superior performance across diverse datasets.