Proceedings of the 6th Computer Science Research Days, JRI 2023, 18-20 December 2023, Ouagadougou, Burkina Faso

Research Article

AI-driven Generation of News Summaries Leveraging GPT and Pegasus Summarizer for Efficient Information Extraction

Download5 downloads
  • @INPROCEEDINGS{10.4108/eai.18-12-2023.2348180,
        author={Issiaka Faissal  Compaore and Rodrigue  Kafando and Aminata  Sabane and Abdoul Kader  Kabore and Tegawende F.  Bissyand\^{e}},
        title={AI-driven Generation of News Summaries Leveraging GPT and Pegasus Summarizer for Efficient Information Extraction},
        proceedings={Proceedings of the 6th Computer Science Research Days, JRI 2023, 18-20 December 2023, Ouagadougou, Burkina Faso},
        publisher={EAI},
        proceedings_a={JRI},
        year={2024},
        month={6},
        keywords={online press artificial intelligence web scraping transformers abstractive summarization},
        doi={10.4108/eai.18-12-2023.2348180}
    }
    
  • Issiaka Faissal Compaore
    Rodrigue Kafando
    Aminata Sabane
    Abdoul Kader Kabore
    Tegawende F. Bissyandé
    Year: 2024
    AI-driven Generation of News Summaries Leveraging GPT and Pegasus Summarizer for Efficient Information Extraction
    JRI
    EAI
    DOI: 10.4108/eai.18-12-2023.2348180
Issiaka Faissal Compaore1,*, Rodrigue Kafando1, Aminata Sabane2, Abdoul Kader Kabore1, Tegawende F. Bissyandé1
  • 1: Centre d’Excellence CITADEL, Université Virtuelle du Burkina Faso
  • 2: Université Joseph KI-ZERBO
*Contact email: issiakafaissal2j@gmail.com

Abstract

The surge of online information makes it challenging to access relevant news quickly. This necessitates automated methods to effectively extract and summarize information. Our research focuses on designing an online press synthesis tool using advanced AI models. We investigate the feasibility of employing two pre-trained models, GPT-3.5 Turbo 16k and Pegasus Summarizer, to generate high-quality summaries from scraped articles. Our methodology encompasses robust web scraping, model integration, and metric evaluation. Experimental results demonstrate that GPT-3.5 Turbo 16k outperforms in accuracy, achieving a BLEU score of 16.39% and a ROUGE score of 0.66%. The turner007/pegasus-summarizer model also performs well, with a BLEU score of 15.45% and a ROUGE score of 0.45%. We identify areas for improvement, such as enriching the database with expert-authored summaries and adopting a dynamic approach to news adaptation. Additionally, we explore a unified models approach for further refinement.