生成式人工智能视角下研究问题与研究方法句生成研究——以高能物理领域为例
Research on the Generation of Research Questions and Method Sentences from the Perspective of AIGC:Taking the Field of High-Energy Physics as an Example
陈昱成 1韩涛1
作者信息
- 1. 中国科学院文献情报中心 北京 100190;中国科学院大学经济与管理学院信息资源管理系 北京 100190
- 折叠
摘要
[研究目的]科研人员从海量的文献中检索与自己研究方向相关的文献需要耗费大量的时间,自动提取研究问题与研究方法句对于学者在研究过程中评估和选择合适的方法具有重要意义.[研究方法]基于多种深度学习分类器构建物理领域研究方向的分类器,筛选高能物理领域的研究,对高能物理领域的论文利用生成式大模型进行研究问题与研究方法句的生成,微调baichuan-7B构建高能物理领域的问答推理工具.[研究结论]研究表明,在研究领域分类中.SciBERT分类结果较好,F1值为82.71％,baichuan-13B在研究问题与研究方法句子生成研究中,2-gram的BLEU值为0.311.最终通过研究问题与研究方法库微调大模型,可实现高能物理领域的问答功能.
Abstract
[Research purpose]It takes a lot of time for researchers to search for literature related to their research direction from a mas-sive amount of literatures.Automatically extracting research questions and method sentences is of great significance for scholars to evaluate and choose appropriate methods during the research process.[Research method]Based on various deep learning classifiers,we construct classifiers for research directions in the field of physics,screen research in the field of high-energy physics,use generative large models to generate research questions and research method sentences for papers in the field of high-energy physics,and fine tune baichuan-7B to construct a question answering reasoning tool in the field of high-energy physics.[Research conclusion]Research has shown that SciB-ERT performs well in research field classification,with an F1 value of 82.71％.Baichuan-13B has a BLEU value of 0.311 for2-gram in research question and research method sentence generation.Ultimately,by fine-tuning the large model through the research question and research method library,the question answering function in the field of high-energy physics can be achieved.
关键词
生成式人工智能/研究方法句/AI4S/高能物理/深度学习Key words
AIGC/research method sentence/AI4S/high Energy Physics/deep learning引用本文复制引用
基金项目
国家社会科学基金项目(22BTQ019)
中国科学院文献情报能力建设专项项目(E329090905)
出版年
2024
国家科技期刊平台
NSTL
万方数据