A Deep Convolutional Neural Network-based Model for Aspect and Polarity Classification in Hausa Movie Reviews

TL;DR

The paper tackles aspect-based sentiment analysis for Hausa movie reviews, a low-resource language, by developing a CNN-based model with attention for simultaneous aspect-term extraction and polarity classification. It creates a Hausa ABSA dataset from YouTube comments (590 entries) and demonstrates that a 9-layer DCNN with embedding, Conv1D, LSTM, and attention achieves high accuracy: $91\%$ on aspect-term extraction and $92\%$ on polarity classification, outperforming traditional machine learning baselines. The work also emphasizes the scarcity of Hausa ABSA resources and provides a practical dataset and methodology to advance cross-cultural linguistic sentiment analysis. A notable limitation is the model’s current handling of a single aspect term per sentence, suggesting transformer-based architectures for multi-aspect analysis in future work.

Abstract

Aspect-based Sentiment Analysis (ABSA) is crucial for understanding sentiment nuances in text, especially across diverse languages and cultures. This paper introduces a novel Deep Convolutional Neural Network (CNN)-based model tailored for aspect and polarity classification in Hausa movie reviews, an underrepresented language in sentiment analysis research. A comprehensive Hausa ABSA dataset is created, filling a significant gap in resource availability. The dataset, preprocessed using sci-kit-learn for TF-IDF transformation, includes manually annotated aspect-level feature ontology words and sentiment polarity assignments. The proposed model combines CNNs with attention mechanisms for aspect-word prediction, leveraging contextual information and sentiment polarities. With 91% accuracy on aspect term extraction and 92% on sentiment polarity classification, the model outperforms traditional machine models, offering insights into specific aspects and sentiments. This study advances ABSA research, particularly in underrepresented languages, with implications for cross-cultural linguistic research.

Figures (5)

• Figure 1: Proposed methodology
• Figure 2: Model architecture
• Figure 3: Distribution of aspect terms and sentiment polarity in the dataset
• Figure 4: Accuracy and loss curves of aspect word extraction
• Figure 5: Accuracy and loss curves of sentiment polarity classification