Abstractive-Summary

Abstractive Text Summarization using Deep Learning (BART)

📌 Project Overview

This project implements a deep learning–based abstractive text summarization system using the BART (Bidirectional and Auto-Regressive Transformer) model. The system automatically generates concise, meaningful, and human-like summaries from long text inputs by understanding context and rephrasing content rather than extracting sentences verbatim.

The project is trained and evaluated using the CNN/DailyMail dataset and provides an interactive web interface built with Gradio for real-time summarization.

🎯 Problem Statement

Large volumes of textual data such as news articles, reports, and documents are difficult to process manually. The objective of this project is to develop an automated summarization system that can:

Understand long paragraphs

Capture key ideas

Generate fluent and generalized summaries using deep learning

💡 Solution Approach

Uses a Transformer-based encoder–decoder architecture

Fine-tunes a pretrained BART model for abstractive summarization

Employs beam search and controlled sampling to generate generalized, paraphrased summaries

Evaluates performance using ROUGE metrics

🧠 Model Architecture

Model: BART (Encoder–Decoder Transformer)

Encoder: Learns contextual representations of input text

Decoder: Generates summary word-by-word

Loss Function: Cross-Entropy Loss

Optimizer: AdamW

Framework: PyTorch

📊 Dataset

Dataset Name: CNN/DailyMail (Version 3.0.0)

Type: News articles with human-written summaries

Usage: Fine-tuning and evaluation of abstractive summarization models

⚙️ Tech Stack Category Tools Programming Language Python Deep Learning PyTorch NLP Library Hugging Face Transformers Dataset Handling Hugging Face Datasets Evaluation ROUGE Score Interface Gradio Environment Google Colab 🚀 Features

Deep learning–based abstractive summarization

Transformer encoder–decoder architecture

Fine-tuning on benchmark dataset

Interactive web interface with:

Text input

Summarize button

Clear button

Save summary option

Generates generalized and paraphrased summaries

🧪 Example

Input Text:

Artificial Intelligence is transforming industries by automating tasks, enhancing decision-making, and enabling innovation across healthcare, finance, and education.

Generated Summary:

AI is reshaping multiple sectors by improving efficiency, supporting smarter decisions, and driving technological innovation.

📈 Evaluation Metrics

The model performance is evaluated using:

ROUGE-1 – Unigram overlap

ROUGE-2 – Bigram overlap

ROUGE-L – Longest common subsequence

Higher ROUGE scores indicate better summary quality.

🖥️ Web Interface

The project includes a Gradio-based UI:

Users can paste long text

Click Summarize to generate output

Clear input instantly

Save summaries for future reference

Output area dynamically expands for readability