Published Paper Details:

Nowadays, a lot of information can be obtained from both online and offline sources. We can access hundreds of documents for a topic. The ability to summarize or create popular topics allows users to quickly search for topics and get initial information as quickly as possible. Manually extracting useful information from them is a difficult task. To solve this problem automatic text summarization (ATS) systems were developed. Text summarization is the process of extracting important information from a large document and turning it into a summary, preserving all the important points. This abstract presents a novel approach using BART pre-trained model which is a large language model used widely for text generation tasks. The proposed system takes input a either a URL, Text or a file of any format. Automatic text summarization aims to produce a shortened version of a long text document while retaining its overall meaning. This has applications like summarizing news articles, research papers, reports, etc. There are two common approaches - extractive methods that identify and extract key sentences from the original text, and abstractive methods which paraphrase and generate new sentences. The proposed system takes input text, documents or URLs and returns a summary. It uses a pretrained BART model fine-tuned on the CNN-Daily Mail dataset. BART (Bidirectional and Auto-Regressive Transformer) combines BERT and GPT architectures. It is a seq2seq model trained as a denoising autoencoder, taking text sequence input and output. Key advantages of BART include improved summary coherence and factual consistency. During training, the model learns to generate a summary given an input document. Current models can produce high quality summaries, but issues remain around coherence, repetition, and accuracy. Key challenges include improving faithfulness to the original text, avoiding repetition, and controlling summary length and style. Future directions involve multi-document summarization, controlled generation, better evaluation metrics, and performance on domain-specific datasets. Overall, text summarization continues to be an important application of NLP that helps condense information and enable quick access to key facts and context. The proposed BART-based system aims to leverage strengths in text generation and compression to improve summarization capabilities.