Citation

Arefin, Md Saiful and Serajun Nabi, Md and Touhami, Meriem and Rehman, Zaka Ur and Ahmad Fauzi, Mohammad Faizal and Abdul Karim, Hezerul (2025) Aspect-Based Sentiment Analysis for Product Reviews using DistilBERT. In: 2025 Multimedia University Engineering Conference, MECON 2025, 21 July 2025 - 23 July 2025, Cyberjaya, Malaysia.

Text IEEE Xplore Full-Text PDF_28.pdf - Published Version Restricted to Repository staff only Download (1MB)

Abstract

Aspect-Based Sentiment Analysis (ABSA) facilitates detailed analysis of customer sentiments by linking sentiments with product attributes. Although large transformer models such as BERT achieve high accuracy in ABSA tasks, their computational cost prevents real-time application in resourcelimited environments. This study examines the performance of DistilBERT, a light and efficient version, for sentiment classification of Amazon mobile-electronics reviews. We present an end-to-end pipeline from tokenization and data preprocessing to model training and testing. The proposed model achieves 91% accuracy and a macro-F1 score of 89.8%, closely comparable to full-size BERT performance while reducing latency and memory viability for real-time ABSA deployment, especially in largescale e-commerce websites. The study also discusses challenges to aspect alignment and proposes improvements in the future through multitask learning and aspect-guided attention.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	DistilBERT, Aspect-Based Sentiment Analysis
Subjects:	H Social Sciences > HF Commerce > HF5001-6182 Business > HF5410-5417.5 Marketing. Distribution of products
Divisions:	Faculty of Artificial Intelligence & Engineering (FAIE)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	17 Mar 2026 07:05
Last Modified:	17 Mar 2026 07:34
URII:	http://shdl.mmu.edu.my/id/eprint/15518

Downloads

Edit (login required)