Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Relational Graph Convolutional Networks for Sentiment Analysis

Asal Khosravi, Zahed Rahmati, Ali Vefghi

TL;DR

This work introduces a heterogeneous text-graph framework that leverages Relational Graph Convolutional Networks (RGCNs) to capture multi-relational dependencies in sentiment analysis. By initializing word and document nodes with pretrained language models (BERT/RoBERTa) and constructing a three-edge-type graph (word-word, word-document, document-document) with PMI, TF-IDF, and Jaccard-based weights, the approach achieves superior performance over traditional GCN-based methods on both English Amazon reviews and Persian Digikala data. The study demonstrates that a two-layer RGCN can effectively propagate relation-specific information, with RoBERTa+RGCN and ParsBERT+RGCN delivering notable gains. The findings highlight the value of relational structure in text graphs and point to future work on dynamic edges, inductive learning, and cross-linguistic extension for practical sentiment analysis applications.

Abstract

With the growth of textual data across online platforms, sentiment analysis has become crucial for extracting insights from user-generated content. While traditional approaches and deep learning models have shown promise, they cannot often capture complex relationships between entities. In this paper, we propose leveraging Relational Graph Convolutional Networks (RGCNs) for sentiment analysis, which offer interpretability and flexibility by capturing dependencies between data points represented as nodes in a graph. We demonstrate the effectiveness of our approach by using pre-trained language models such as BERT and RoBERTa with RGCN architecture on product reviews from Amazon and Digikala datasets and evaluating the results. Our experiments highlight the effectiveness of RGCNs in capturing relational information for sentiment analysis tasks.

Relational Graph Convolutional Networks for Sentiment Analysis

TL;DR

This work introduces a heterogeneous text-graph framework that leverages Relational Graph Convolutional Networks (RGCNs) to capture multi-relational dependencies in sentiment analysis. By initializing word and document nodes with pretrained language models (BERT/RoBERTa) and constructing a three-edge-type graph (word-word, word-document, document-document) with PMI, TF-IDF, and Jaccard-based weights, the approach achieves superior performance over traditional GCN-based methods on both English Amazon reviews and Persian Digikala data. The study demonstrates that a two-layer RGCN can effectively propagate relation-specific information, with RoBERTa+RGCN and ParsBERT+RGCN delivering notable gains. The findings highlight the value of relational structure in text graphs and point to future work on dynamic edges, inductive learning, and cross-linguistic extension for practical sentiment analysis applications.

Abstract

With the growth of textual data across online platforms, sentiment analysis has become crucial for extracting insights from user-generated content. While traditional approaches and deep learning models have shown promise, they cannot often capture complex relationships between entities. In this paper, we propose leveraging Relational Graph Convolutional Networks (RGCNs) for sentiment analysis, which offer interpretability and flexibility by capturing dependencies between data points represented as nodes in a graph. We demonstrate the effectiveness of our approach by using pre-trained language models such as BERT and RoBERTa with RGCN architecture on product reviews from Amazon and Digikala datasets and evaluating the results. Our experiments highlight the effectiveness of RGCNs in capturing relational information for sentiment analysis tasks.
Paper Structure (24 sections, 9 equations, 6 figures, 4 tables)

This paper contains 24 sections, 9 equations, 6 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Proposed Method
  4. Text Pre-processing
  5. Graph Construction
  6. Heterogeneous graph
  7. Creating edges
  8. Pre-trained language models
  9. RGCN framework
  10. Details of our proposed method
  11. Training settings
  12. The model
  13. Experiments
  14. Datasets
  15. The Amazon dataset
  16. ...and 9 more sections

Figures (6)

  • Figure 1: Overview of the proposed method
  • Figure 2: Overview of the constructed graph - The constructed graph is a directed and weighted heterogeneous graph with two types of nodes (words and documents) and three types of edges ( word-word, word-document, and document-document )
  • Figure 3: Overview of RGCN and its corresponding computational graph for the example node "A".
  • Figure 4: The computation view for updating a single graph node in the RGCN model
  • Figure 5: (a) distribution of labels within comments for the dataset with 5 classes (b) distribution of labels within comments for the dataset with 2 classes
  • ...and 1 more figures