View Proposal

---

Proposer

Timothy Yap

Title

Investigating Cross-Domain Transfer Learning for Aspect-Based Sentiment Analysis

Goal

To investigate the effectiveness of transfer learning techniques for improving the generalization of Aspect-Based Sentiment Analysis (ABSA) models across different application domains.

Description

Aspect-Based Sentiment Analysis (ABSA) aims to identify sentiment expressed towards specific aspects of an entity, such as product features, services, or customer experiences. While recent machine learning and deep learning approaches have achieved promising results, many models are trained and evaluated within a single domain and often experience performance degradation when applied to unseen domains. This project aims to investigate the effectiveness of cross-domain transfer learning for ABSA. The study will examine whether knowledge learned from one domain can be transferred to another domain with limited labeled data. The student will develop and evaluate ABSA models using datasets from multiple domains, such as restaurant reviews, laptop reviews, hotel reviews, or product reviews. The project will compare different transfer learning strategies, including direct transfer, fine-tuning, and domain adaptation approaches. Experimental results will be analyzed to understand the factors affecting cross-domain performance and the robustness of various models in real-world deployment scenarios. The findings may contribute to the development of more generalizable ABSA systems that require less domain-specific training data. Research Questions: How well do ABSA models generalize across different domains? Does fine-tuning improve cross-domain performance? Which transfer learning strategy provides the best balance between accuracy and training cost? What aspects are most sensitive to domain shifts?

Resources

Python, Hugging Face Transformers, PyTorch, Scikit-learn, Aspect-Based Sentiment Analysis Datasets (SemEval and related datasets)

Background

Python Programming, Machine Learning, Natural Language Processing, Deep Learning Fundamentals

Url

External Link

Difficulty Level

Variable

Ethical Approval

Full

Number Of Students

1

Supervisor

Timothy Yap

Keywords

aspect-based sentiment analysis, transfer learning, domain adaptation, natural language processing, transformers, machine learning

Degrees

Bachelor of Science in Computing Science