Enhancing Performance of Abstractive Multi-Document Update Summarization on TAC Dataset

Abstract

In this paper, we investigate the efficacy of various cutting-edge models for update summarization on the TAC 2009 dataset. To construct abstractive and extractive summaries of news items, we use the T5 Transformer model and Textrank + Pegasus model. Our goal is to assess how well these models capture key information from updates and generate coherent and useful summaries. Here we use conventional assessment measures such as ROUGE to assess the performance of the models. We analyze the fluency, coherence, and informativeness of generated summaries from the T5 Transformer model, Textrank + Pegasus, and TensorFlow models against human-authored gold summaries.