Modeling Structured Dependency Tree with Graph Convolutional Networks for Aspect-Level Sentiment Classification | JoVE Visualize

Area of Science:

Natural Language Processing
Artificial Intelligence
Machine Learning

Background:

Aspect-based sentiment analysis requires fine-grained prediction of sentiment polarities for specific aspects within sentences.
Current graph neural network models often use dependency trees but suffer from noisy nodes and fail to capture crucial indirect relationships.
Existing methods overlook connections between nodes without direct dependency edges that significantly influence sentiment polarity.

Purpose of the Study:

To propose a novel Structured Dependency Tree-based Graph Convolutional Network (SDTGCN) model.
To address limitations in existing models by improving the capture of contextual relationships and sentiment dependencies.
To enhance the precision of aspect representations in sentiment analysis.

Main Methods:

Constructing a structured syntactic dependency graph incorporating positional information, sentiment commonsense knowledge, part-of-speech tags, and dependency distances.
Assigning arbitrary edge weights to enhance connections between aspect nodes and pivotal words while weakening irrelevant links.
Utilizing part-of-speech tags and dependency distances to identify relationships between nodes without direct dependencies.
Aggregating node information based on importance to derive precise aspect representations.

Main Results:

The SDTGCN model demonstrated superiority over state-of-the-art approaches on five publicly available datasets.
Significant improvements were observed in accuracy and F1-score on the majority of datasets, with increases up to 1.17.
The model effectively enhances sentiment classification performance by better expressing sentiment dependencies.

Conclusions:

The proposed SDTGCN model offers a significant advancement in aspect-based sentiment analysis.
Incorporating structured syntactic information and commonsense knowledge improves the model's ability to capture nuanced sentiment.
The enhanced performance validates the effectiveness of the SDTGCN approach for precise sentiment classification.