Exploring Extractive and Abstractive Approaches for Myanmar Language Text Summarization

Text summarization is one of the most difficult natural language processing (NLP) tasks. It is the process of creating concise, intelligent summaries from a variety of sources, including books, news articles, and social media posts. This method saves readers time in today's busy environment while giving them fast access to important information. This paper presents extractive and abstractive summarization approaches for Myanmar text. While abstractive summarizing creates the complete summary from scratch using the input text, extractive summarization chooses a subset of sentences from the original text for the summary. Abstractive summarizing is more sophisticated than extractive summarization since it involves human-like elements to create the summary from the ground up. The TextRank algorithm and cosine similarity were used for extractive summarization. The mT-5 model, a transformer-based seq2seq model, was employed to carry out abstractive summarization on the XL-Sum dataset. As a result, the extractive summarization of the proposed system yields 43.14 of the ROUGE-1 score, while the abstractive summarization contributes 36.67 of the ROUGE-1 score.