Scholarlink

Over the years, the literature in the field of biomedical science has increased exponentially. Structured literature helps readers to grasp the content efficiently. Segmenting abstract sentences into categories like background, objective, method, results, and conclusions will benefit readers and will aid information extraction and information retrieval in complex biomedical literature. This paper focuses on randomized clinical trials (RCTs) and presents a comparative study of fine-tuned BERT-based variants for move recognition on the RCMR_RCT dataset, a subset of RCMR 280k. Among the models used in the experiment, BioBERT has outperformed BERT, SciBERT, PubMedBERT, BioMedBERT, BART, and RoBERTa. Utilizing the RCMR_RCT dataset instead of the PubMed 20k RCT dataset using BERT, BioBERT, SciBERT, and PubMedBERTmodel demonstrates an improvement of 4.95%, 5.71%, 3.95%, and 4.32% in the F1 score metric. This paper contributes to advancing the biomedical literature analysis by developing methods for move recognition.

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

Harnessing Pre-trained Language Models for Efficient Move Recognition in Biomedical Abstracts

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Gaurav Patil,
  • Robert E. Mercer

摘要

Over the years, the literature in the field of biomedical science has increased exponentially. Structured literature helps readers to grasp the content efficiently. Segmenting abstract sentences into categories like background, objective, method, results, and conclusions will benefit readers and will aid information extraction and information retrieval in complex biomedical literature. This paper focuses on randomized clinical trials (RCTs) and presents a comparative study of fine-tuned BERT-based variants for move recognition on the RCMR_RCT dataset, a subset of RCMR 280k. Among the models used in the experiment, BioBERT has outperformed BERT, SciBERT, PubMedBERT, BioMedBERT, BART, and RoBERTa. Utilizing the RCMR_RCT dataset instead of the PubMed 20k RCT dataset using BERT, BioBERT, SciBERT, and PubMedBERTmodel demonstrates an improvement of 4.95%, 5.71%, 3.95%, and 4.32% in the F1 score metric. This paper contributes to advancing the biomedical literature analysis by developing methods for move recognition.