Abstract: The exponential growth of online textual data has created a critical need for efficient information processing, making automated text summarization an indispensable tool for mitigating information overload. While traditional methods have struggled with fluency and coherence, the advent of Transformer-based models has defined a new state-of-the-art. This paper introduces a robust, end-to-end framework for abstractive text summarization leveraging a pre-trained BART (Bidirectional and Auto-Regressive Transformers) model. The system utilizes the facebook/bart-large-cnn model, a specific variant fine-tuned on the CNN/Daily Mail news dataset, which employs a bidirectional encoder for comprehension and an auto-regressive decoder to generate novel text. This AI core is deployed within a modern, scalable web application, served via a high-performance FastAPI backend API and consumed by an interactive React user interface. This paper details the full-stack architecture, from the model-loading strategy at server startup to the asynchronous API request-response workflow. The model's performance is quantitatively evaluated using the standard ROUGE metrics, demonstrating strong results with a mean ROUGE-1 F1 score of 50.61% and a ROUGE-L F1 score of 42.99%. We provide a detailed analysis of these metrics, including precision/recall trade-offs and score distributions, confirming the model's high abstractive capability. This research serves as a comprehensive blueprint for the practical implementation and evaluation of a state-of-the-art Transformer approach for real-world summarization applications.

Keywords: Abstractive Text Summarization, BART, Transformer, Natural Language Processing (NLP), ROUGE, FastAPI, React, Full-Stack Application, CNN/Daily Mail.

Downloads: | DOI: 10.17148/IJIREEICE.2025.131035