Authors:
Santosh Baral, Saroj Basyal, Shikha Jha, Shirsti Sharma, Mutaher Ahmed

Addresses:
Department of Computer Science and Technology, University of the West of Scotland, Glasgow, Scotland, United Kingdom. School of Computing and Artificial Intelligence, University of Law, London, England, United Kingdom.

Abstract:

Customer reviews contain valuable information about customer experiences, product quality, and satisfaction. However, the increasing volume of online reviews makes manual analysis impossible. In this work, researchers design the Explainable Agentic AI for Automated Customer Insight Generation from Product Reviews system, which integrates NLP, sentiment analysis, topic modeling, XAI, and evidence-based customer insight generation into a unified analytics approach. The system was built in Python using the Jupyter Notebook framework and tested on an Amazon product review dataset of 72,500 reviews across five product categories. Researchers preprocessed and classified customer reviews using VADER, Logistic Regression, and Naïve Bayes. They used Latent Dirichlet Allocation (LDA) to identify discussion topics. To improve model transparency and explain sentiment predictions, researchers used SHAP and LIME. A deterministic, template-driven evidence synthesis process has generated structured customer insight reports from sentiment, topic, and explainability outputs. Researchers also implemented an interactive visualization dashboard. As a result of the experiment, Naive Bayes achieved the highest overall accuracy, while Logistic Regression achieved the highest Macro F1-score and was selected for the Explainable AI stage.

Keywords: Agentic AI; Sentiment Analysis; Customer Insight Generation; Five Product Categories; Latent Dirichlet Allocation; Explainable Agentic AI; Jupyter Notebook framework.

Received on: 21/12/2025, Revised on: 04/02/2025, Accepted on: 21/03/2026, Published on: 16/10/2026

DOI: 10.69888/FTSFDS.2026.000803

FMDB Transactions on Sustainable Finance and Data Science, 2026 Vol. 1 No. 4, Pages: 240-267

• Views : 59
• Downloads : 7

Download PDF