VIETNAMESE ABSTRACTIVE TEXT SUMMARIZATION USING PROPHETNET

Abstract

The thesis studies a method of creating abstractive text summaries using pre trained ProphetNet model in Vietnamese language. In our experiment, the pretrained model will be fine-tuned on a Vietnamese dataset. During inference, Beam Search algorithm is used to select words generated for summaries. The summary which the model generated should consist of sentences which represent the core ideas of the Vietnamese original documents. ROUGE metric is used to evaluate the performance of the abstractive text summarization model.