Scholarlink

TextSumEval aim to provide a platform for the researcher working the field of text summarization and evaluation. In this task, we have provided the system-generated summaries and gold summaries to the task participants. Participants need to design a summary evaluation technique that compares system-generated summaries with gold summaries and generated their semantic resemblance score. To facilitate the above-mentioned task, we have provided a total of 6000 summaries which are selected from the existing databases, namely, CNNDM, MSMO, and XSUM. We have calculated the correlation scores between the BLEU score, ROUGE score, and our baseline system’s semantic score with participants submitted semantic scores using a Pearson correlation coefficient to verify the participants’ scores. Moreover, based on the mean of the correlation scores, we have highlighted the methodology of the best-performed team.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

TextSumEval-2022: Overview of the Textual Summary Evaluation Task

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Tamwo,
  • Pankaj Dadure,
  • Prottay Kumar Adhikary,
  • Partha Pakray,
  • Benoit Favre,
  • Sivaji Bandyopadhyay,
  • Thierry Artires

摘要

TextSumEval aim to provide a platform for the researcher working the field of text summarization and evaluation. In this task, we have provided the system-generated summaries and gold summaries to the task participants. Participants need to design a summary evaluation technique that compares system-generated summaries with gold summaries and generated their semantic resemblance score. To facilitate the above-mentioned task, we have provided a total of 6000 summaries which are selected from the existing databases, namely, CNNDM, MSMO, and XSUM. We have calculated the correlation scores between the BLEU score, ROUGE score, and our baseline system’s semantic score with participants submitted semantic scores using a Pearson correlation coefficient to verify the participants’ scores. Moreover, based on the mean of the correlation scores, we have highlighted the methodology of the best-performed team.