Advancements in the Efficacy of Flan-T5 for Abstractive Text Summarization: A Multi-Dataset Evaluation Using ROUGE and BERTScore
Abdulrahman Mohsen Ahmed Zeyad, Arun Biradar · 2024
This research ventured into the realm of abstractive text summarization, focusing on the amalgamation and efficacy of sophisticated NLP models, notably Flan-T5. We employed these cutting-edge models on a variety of datasets, such as XSum, CNN/DailyMail, Multi-News, Newsroom, and Gigaword, to gauge their summarization abilities. The models' performance was assessed using complex metrics like ROUGE and BERTScore. A remarkable discovery was the attainment of a ROUGE-L score of 0.5021 on the Gigaword dataset, underscoring the models' proficiency in producing coherent and contextually precise summaries. The outcomes were substantial, indicating a noticeable improvement in the quality, coherence, and contextual accuracy of the summaries generated by these models. This study concludes that the application of Flan-T5 signifies a considerable progression in the field of abstractive text summarization. Their capacity to effectively process and abridge extensive information is a reflection of their technological capability and constitutes a significant step forward in NLP, opening up new avenues for data processing and knowledge dissemination.