Title: Quality Assessment Using Large Language Models and Prompt Engineering

Authors: Mukesh Kumar A, Gokulnath G, Mrs. A. Jeyanthi

Abstract: We present a novel and intelligent framework that integrates web scraping, domain-specific sentiment classification, and Large Language Models (LLMs) to evaluate product quality based on customer reviews in the e-commerce domain. The proposed system is designed to be fully automated and scalable, capable of extracting and analyzing user-generated content from popular platforms such as Flipkart. Using customized web scraping modules, the system collects real-time reviews and filters them through a preprocessing pipeline. These reviews are then contextually categorized into predefined product domains such as durability, comfort, performance, and usability, thereby enabling a more granular understanding of customer sentiment. For sentiment classification, we employ DeepSeek-R1, a state-of-the-art open-source LLM hosted and accelerated on Groq— a high-speed cloud infrastructure optimized for inference workloads. This allows for efficient and context-aware sentiment analysis that outperforms traditional approaches in terms of speed, scalability, and accuracy. The sentiment outcomes are further transformed into structured data representations, which are visualized through interactive dashboards built using PyQt5 and Matplotlib. These dashboards support real-time filtering, trend analysis, and comparison across products, offering stakeholders actionable insights into customer satisfaction. The system was empirically evaluated on a manually annotated dataset comprising 50 reviews across three product categories. The classification pipeline achieved an accuracy of 98%, precision of 0.99, recall of 0.97, and an F1 score of 0.98. Additionally, a Cohen’s Kappa score of 0.953 was recorded, indicating near-perfect agreement with human annotations. These results demonstrate the robustness and reliability of our LLM-based approach in real-world e-commerce scenarios, setting a new benchmark for intelligent review analysis systems.

DOI: http://doi.org/