Automatic Multi-Document Text Summarization

Abstract

This research explores automatic multi-document text summarization as a solution to automate the summarisstation of substantial amounts of online data. The goal is to produce succinct summaries that effectively communicate the primary idea of published works without using redundant language.This study aims to thoroughly explore the strengths and weaknesses of the Flan-T5 language model for text summarization, using the samsum dataset as a benchmark for dialogue summarization tasks. Through systematic experimentation varying hyperparameters, fine-tuning approaches, and training conditions, we will provide comprehensive insight into the model’s capabilities and limitations. By testing Flan-T5 under diverse scenarios, we will uncover how factors like batch size, learning rate, and dataset size affect its ability to extract key information from source texts. Our research will elucidate the complex interplay between hyperparameters and summary quality, underscoring their importance in optimizing summarization performance. Overall, this investigation will deliver a nuanced understanding of how the Flan-T5 model performs on dialogue summarization, highlighting avenues for improvement.