T5, BERT and Seq2Seq model evaluation based on Rouge and BLEU scores

Project information

  • Category: Text Summarization (Generative)
  • Project date: 8 December, 2023
  • Project URL: : Project Link
  • Paper: Link

Project Description

  • Objective: Automatically generating concise and informative headlines from extensive news articles, employing various advanced NLP models and techniques to summarize content effectively​
  • Conducted training sessions with limited data to enable T5 for headline generation, capitalizing on its unique architecture.
  • Employed T5, Bert, and Seq2Seq models, inspired by the 2019 ACL paper, for Text Summarization, generating article headlines by leveraging distinct approaches
  • Utilized pre-trained BERT models specified in the "Text Summarization with pre-trained encoders" paper to produce text summaries, harnessing their established capabilities.
  • Developed and trained the Seq2Seq model from the ground up, enabling comprehensive understanding and control over its training process and outcomes.
  • Conducted a thorough comparative analysis of the models' performances, evaluating against Blue score and Rouge score metrics, with detailed findings consolidated in a comprehensive research paper.