[Paper Review] Arabic aspect based sentiment analysis using bidirectional GRU based models

Mohammed M. Abdelgwad, Taysir Hassan A. Soliman+2 more|arXiv (Cornell University)|Jan 22, 2021

Sentiment Analysis and Opinion Mining79 references103 citations

TL;DR

This paper proposes two bidirectional GRU-based deep learning models for Arabic aspect-based sentiment analysis: BGRU-CNN-CRF for opinion target extraction (T2) and IAN-BGRU for aspect sentiment polarity classification (T3). The IAN-BGRU model achieves state-of-the-art performance with 83.98% accuracy on T3 and a 39.7% F1-score improvement on T2 over baseline methods using the SemEval-2016 Arabic hotel reviews dataset.

ABSTRACT

Aspect-based Sentiment analysis (ABSA) accomplishes a fine-grained analysis that defines the aspects of a given document or sentence and the sentiments conveyed regarding each aspect. This level of analysis is the most detailed version that is capable of exploring the nuanced viewpoints of the reviews. The bulk of study in ABSA focuses on English with very little work available in Arabic. Most previous work in Arabic has been based on regular methods of machine learning that mainly depends on a group of rare resources and tools for analyzing and processing Arabic content such as lexicons, but the lack of those resources presents another challenge. In order to address these challenges, Deep Learning (DL)-based methods are proposed using two models based on Gated Recurrent Units (GRU) neural networks for ABSA. The first is a DL model that takes advantage of word and character representations by combining bidirectional GRU, Convolutional Neural Network (CNN), and Conditional Random Field (CRF) making up the (BGRU-CNN-CRF) model to extract the main opinionated aspects (OTE). The second is an interactive attention network based on bidirectional GRU (IAN-BGRU) to identify sentiment polarity toward extracted aspects. We evaluated our models using the benchmarked Arabic hotel reviews dataset. The results indicate that the proposed methods are better than baseline research on both tasks having 39.7% enhancement in F1-score for opinion target extraction (T2) and 7.58% in accuracy for aspect-based sentiment polarity classification (T3). Achieving F1 score of 70.67% for T2, and accuracy of 83.98% for T3.

Motivation & Objective

• Address the lack of advanced Arabic ABSA methods by developing deep learning models tailored for Arabic.
• Improve opinion target extraction (T2) and aspect sentiment polarity classification (T3) in Arabic hotel reviews.
• Overcome limitations of traditional machine learning approaches that rely on scarce Arabic NLP resources.
• Utilize bidirectional GRUs and attention mechanisms to enhance contextual modeling without external lexicons.
• Achieve state-of-the-art performance on the SemEval-2016 Arabic ABSA benchmark dataset.

Proposed method

• Employs a BGRU-CNN-CRF model combining bidirectional GRUs, convolutional layers, and CRF for sequence labeling in opinion target extraction (T2).
• Uses word and character-level embeddings as input to capture morphological and semantic features in Arabic.
• Applies an interactive attention network (IAN) with bidirectional GRUs to model aspect terms and their contexts interactively.
• Integrates attention mechanisms to dynamically focus on relevant words for sentiment classification, improving representation learning.
• Trains models end-to-end using backpropagation with cross-entropy loss, optimizing for both T2 and T3 tasks.
• Evaluates models on the SemEval-2016 Arabic hotel reviews dataset, using F1-score and accuracy as metrics.

Experimental results

Research questions

• RQ1 Can bidirectional GRU-based models outperform traditional machine learning baselines in Arabic ABSA tasks?
• RQ2 How effective is the IAN-BGRU architecture in capturing aspect-sentiment dependencies in Arabic reviews?
• RQ3 To what extent do word and character-level embeddings improve opinion target extraction in low-resource Arabic NLP?
• RQ4 Does the use of attention mechanisms reduce reliance on external lexicons in Arabic sentiment analysis?
• RQ5 Can the BGRU-CNN-CRF model achieve robust opinion target extraction without dependency on handcrafted features?

Key findings

• The IAN-BGRU model achieved 83.98% accuracy on aspect sentiment polarity classification (T3), outperforming the best baseline by 7.58 percentage points.
• The BGRU-CNN-CRF model achieved a 70.67% F1-score on opinion target extraction (T2), representing a 39.7% improvement over the baseline.
• The IAN-BGRU model outperformed all other variants, including IAN-LSTM and IAN-GRU, due to enhanced bidirectional context modeling.
• The use of bidirectional GRUs improved performance over unidirectional models by capturing long-range dependencies in both forward and backward directions.
• The model successfully classified conflicting sentiments in a single sentence—e.g., positive for location and negative for rooms—without external lexicons.
• The ablation study confirmed that attention mechanisms and bidirectional modeling significantly enhance performance, while complex models like MEMNET did not yield improvements.