Authors:
Prem Kumar Mehta, RejwanBin Sulaiman
Addresses:
Department of Computer Science and Technology, University of the West of Scotland, Paisley, Scotland, United Kingdom.
Abstract:
This study examines how machine learning, deep learning, and transformer-based techniques can be applied to classify sentiment in customer reviews and generate short, abstractive summaries from them. The research uses a cleaned dataset of genuine customer feedback, including both brief comments and longer narrative entries. Classical machine-learning models were first applied to create a baseline, followed by neural-network models, and finally transformer architectures, such as BERT for sentiment analysis and PEGASUS for summarisation. All models were trained and tested on the same processed dataset to maintain fairness in comparison. The results show that classical methods handle simple, straightforward reviews well, while deep learning models handle longer, more descriptive text more effectively. Across both tasks, the transformer models produced the strongest and most consistent performance. BERT achieved the highest accuracy in sentiment classification, and PEGASUS generated clear summaries that captured the main ideas of the reviews. Taken together, the findings suggest that transformer-based models are better suited to analyzing informal customer feedback that varies in structure and often carries mixed or complex emotions.
Keywords: Deep Learning (DL); Transformer Models; Text Summarisation; Customer Reviews; Bert and Pegasus; Sentiment Analysis; Abstractive Summarisation; Complex Emotions.
Received on: 28/07/2025, Revised on: 19/09/2025, Accepted on: 07/10/2025, Published on: 05/03/2026
DOI: 10.64091/ATICR.2026.000309
AVE Trends in Intelligent Computing Research, 2026 Vol. 1 No. 1 , Pages: 35-64