图书情报工作

图书情报工作 >

2025 , Vol. 69 >Issue 17: 72 - 83

DOI: https://doi.org/10.13266/j.issn.0252-3116.2025.17.007

研究论文

大模型辅助发现新颖科研问题的方法

  • 韩红旗 ,
  • 张均胜 ,
  • 周则旭 ,
  • 吴光
展开
  • 1 中国科学技术信息研究所 北京 100038;
    2 富媒体数字出版内容组织与知识服务重点实验室 北京 100038
韩红旗，研究员，博士，E-mail：bithhq@163.com；张均胜，研究员，博士；周则旭，硕士研究生；吴光，硕士研究生。

收稿日期: 2024-08-02

  修回日期: 2024-11-23

  网络出版日期: 2025-08-20

基金资助

本文系国家科技重大专项“高可信科技文献预训练大模型构建与优化”（项目编号：2023ZD0121502）和中国科学技术信息研究所创新基金面上项目“基于文本生成模型的问题创新方案挖掘”（项目编号：MS2024-03）研究成果之一。

收起

Discovering Novel Research Problems Using Text Generation Technology Based on Large Language Model

  • Han Hongqi ,
  • Zhang Junsheng ,
  • Zhou Zexu ,
  • Wu Guang
Expand
  • 1 Institute of Scientific and Technical Information of China, Beijing 100038;
    2 Key Laboratory of Rich-Media Knowledge Organization and Service of Digital Publishing Content, NPPA (National Press and Publication Administration), Beijing 100038
Han Hongqi, Research Fellow, PhD, E-mail: bithhq@163.com; Zhang Junsheng, Research Fellow, PhD; Zhou Zexu, Master’s Candidate; Wu Guang, Master’s Candidate.

Received date: 2024-08-02

  Revised date: 2024-11-23

  Online published: 2025-08-20

Supported by

This work is supported by National Science and Technology Major Project, titled “Construction and Optimization of Pre-Trained Large Models for Highly Credible Scientific Literature” (Grant No. 2023ZD0121502) and the General Projects of ISTIC Innovation Foundation, titled “Problem Innovation Solution Mining Based on Text Generation Model” (Grant No. MS2024-03).

Fold

摘要

[目的/意义] 科学研究从问题开始，问题推动研究、引导研究。由计算机生成新颖的科研问题，有利于激发科研人员的创造性思维和创新灵感，辅助科研人员提出新的科研问题。[方法/过程] 借鉴基于片段预测的词汇约束文本生成方法，提出一个基于大模型的科研问题生成方法，能根据用户输入的关键词生成新的科研问题，其生成过程包括问题句子抽取、训练数据构造、大模型微调、生成文本处理和科研问题选择等5个环节。[结果/结论] 选择通义千问Qwen2-1.5B-Instruct作为基座模型进行实验研究，采用约束关键词比例、文本相似性、文本多样性、文本流利度4项指标评估生成文本的效果，采用人工方法评估生成问题的新颖性，并将生成的问题与论文中新提出的问题进行对比，以说明科研问题生成的价值。实验发现，提出的方法能生成可理解的、新颖的、具有参考价值的科研问题；与Bart模型对比发现，大模型（Qwen2）生成的问题文本能很好地包含约束关键词，且在文本多样性方面表现突出。

关键词： 科研问题; 文本生成; 大模型; LoRA微调; 预训练模型; 通义千问

本文引用格式

韩红旗 , 张均胜 , 周则旭 , 吴光 . 大模型辅助发现新颖科研问题的方法[J]. 图书情报工作, 2025 , 69(17) : 72 -83 . DOI: 10.13266/j.issn.0252-3116.2025.17.007

Abstract

[Purpose/Significance] Scientific research starts with problems, which drive and guide research. The novel scientific research problems generated by computers can stimulate the creative thinking and innovative inspiration of researchers, and assist them in proposing new scientific research problems. [Method/Process] Drawing on the method of the lexically constrained text generation based on segment predication (LCTG-SP), a research problem generation method based on a LLM was proposed, which could generate new research problems based on user input keywords. The generation process included five steps: problem sentence extraction, training data construction, large model fine-tuning, generated text processing, and research problem selection. [Result/Conclusion] The Qwen2-1.5B-Instruction model provided by TongyiQianwen is selected as the base LLM for experimental research, and four indicators including constrained keyword ratio, text similarity, text diversity, and text fluency are used to evaluate the effectiveness of generated text. The experiments find that the proposed method can generate understandable and novel scientific research problems. Compared with the Bart model, it finds that the text generated by the LLM (Qwen2) can well contain constrained keywords and perform outstandingly in terms of text diversity. By manually evaluating the generated scientific research problems and comparing them with the author’s newly proposed problems, it concludes that they have novelty and reference value, which can stimulate researchers’ creative thinking.

Key words： scientific research problem; text generationlarge; language model; LoRA fine-tuning; pre-trained models; Tongyi Qianwen

参考文献

[1] 林定夷. 科学问题与科学目标[J]. 中国社会科学, 1991(5): 29-38. (LIN D Y. Scientific problems and scientific objectives[J]. Social sciences in China, 1991(5): 29-38.)
[2] 索传军, 赖海媚. 学术论文问题知识元的类型与描述规则[J]. 中国图书馆学报, 2021, 47(2): 95-109. (SUO C J, LAI H M. Types and description rules of problem knowledge units in academic papers[J]. Journal of library science in China, 2021, 47(2): 95-109.)
[3] PARK M, LEAHEY E, FUNK R. Papers and patents are becoming less disruptive over time[J]. Nature, 2023(613): 138–144.
[4] LEWIS M, LIU Y, GOYAL N, et al. BART: denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension[C]//Proceedings of the 58th annual meeting of the Association for Computational Linguistics. Seattle: ACL, 2020: 7871-7880.
[5] 桑基韬, 于剑. 从ChatGPT看AI未来趋势和挑战[J]. 计算机研究与发展, 2023, 60(6): 1191-1201. (SANG J T, YU J. ChatGPT: aglimpse into AI's future [J]. Journal of computer research and development, 2023, 60(6): 1191-1201.)
[6] 聂锦燃, 杨麟儿, 杨尔弘. 基于片段预测的词汇约束文本生成[J]. 中文信息学报, 2023, 37(8): 150-158. (NIE J R, YANG L E, YANG E H. Lexically constrained text generation basedon segments prediction[J]. Journal of Chinese information processing, 2023, 37(8): 150-158.)
[7] MIKOLOV T, CHEN K, CORRADO G, et al. Efficient estimation of word representations in vector space[EB/OL]. [2025-04-01]. https://doi.org/10.48550/arXiv.1301.3781.
[8] 李敏, 张怡, 王善玲. 基于Word2vec及TF-MONO算法的中国观众对中外电影的需求与满意度比较研究[J]. 情报工程, 2022, 8(2): 73-86. (LI M, ZHANG Y, WANG S L. A comparative study of Chinese audience's demand and satisfaction for Chinese and foreign movies based on word2vec and TF-MONO[J]. Technology intelligence engineering, 2022, 8(2): 73-86.)
[9] 温书豪, 黄学军, 周良, 等. 刑事案件上诉原因文本挖掘——基于二审文书数据[J]. 情报工程, 2021, 7(1): 103-114. (WEN S H, HUANG X J, ZHOU L. The influencing factors analysis of second criminal instance based on text mining[J]. Technology intelligence engineering, 2021, 7(1): 103-114.)
[10] PETERS M, NEUMANN M, IYYER M, et al. Deep contextualized word representations[C]//Proceedings of NAACL-HLT 2018. Louisiana: ACL, 2018: 2227-2237.
[11] 孙凯丽, 罗旭东, 罗有容. 预训练语言模型的应用综述[J]. 计算机科学, 2023, 50(1): 176-184. (SUN K L, LUO X D, LUO Y R. Survey of applications of pretrained language models[J]. Computer science, 2023, 50(1): 176-184.)
[12] 周强伟, 施水才, 王洪俊. 基于预训练模型的受控文本生成研究综述[J]. 软件导刊, 2024, 23(4): 199-207. (ZHOU Q W, SHI S C, WANG H J. Overview of controlled text generation based on pre-trained models[J]. Software guide, 2024, 23(4): 199-207.)
[13] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]//Proceedings of the 31st international conference on neural information processing systems. Long Beach: ACM, 2017: 6000-6010.
[14] DEVLIN J, CHANG M, LEE K, et al. BERT: Pre-training of deep bidirectional transformers for language understanding[C]//Proceedings of the 2019 conference of the North American Chapter of the Association for Computational Linguistics: human language technologies, Minneapolis: ACL, 2019: 4171-4186.
[15] OUYANG L, WU J, XU J, et al. Training language models to follow instructions with human feedback[C]//Proceedings of the 36th international conference on neural information processing system. New York: ACM, 2022.
[16] 刘睿珩, 叶霞, 岳增营. 面向自然语言处理任务的预训练模型综述[J]. 计算机应用, 2021, 41(5): 1236-1246. (LIU R H, YE X, YUE Z Y. Review of pre-trained models for natural language processing tasks[J]. Journal of computer applications, 2021, 41(5): 1236-1246.)
[17] ZHAO W, ZHOU K, LI J, et al. A survey of large language models[EB/OL]. [2025-04-01]. https://doi.org/10.48550/arXiv.2303.18223.
[18] MINAEE S, MIKOLOV T, NIKZAD N, et al. Large language models: a survey[EB/OL]. [2025-04-01]. https://doi.org/10.48550/arXiv.2402.06196.
[19] 丁鑫, 邹荣金, 潘志庚. 基于高效参数微调的生成式大模型领域适配技术[J]. 人工智能, 2023(4): 1-9. (DING X, ZOU R J, PAN Z G. Domain adaptation technology for generative large models based on efficient parameter fine-tuning[J]. AI-View, 2023(4): 1-9.)
[20] 苏劲松, 陈骏轩, 陆紫耀, 等. 篇章神经机器翻译综述[J]. 情报工程, 2020, 6(5): 4-14. (SU J S, CHEN J X, LU Z Y, et al. A Survey of document-level neural machine translation[J]. Technology intelligence engineering, 2020, 6(5): 4-14.)
[21] MULLA N, GHARPURE P. Automatic question generation: a review of methodologies, datasets, evaluation metrics, and applications[J]. Progress in artificial intelligence, 2023, 12: 1–32.
[22] 吴云芳, 张仰森. 问题生成研究综述[J]. 中文信息学报, 2021, 35(7): 1-9. (WU Y F, ZHANG Y S. A survey of question generation[J]. Journal of Chinese information processing, 2021, 35(7): 1-9.)
[23] 邱嘉作, 熊德意. 神经问题生成前沿综述[J]. 计算机科学, 2021, 48(6): 159-167. (QIU J Z, XIONG D Y. Frontiers in neural question generation: aliterature review[J]. Computer science, 2021, 48(6): 159-167.)
[24] KURDI G, LEO J, PARSIA B, et al. A systematic review of automatic question generation for educational purposes[J]. International journal of artificial intelligence in education, 2020, 30: 121-204.
[25] LU C, LU S. A survey of approaches to automatic question generation: from 2019 to early 2021[C]//Proceedings of the 33rd conference on computational linguistics and speech processing. Taoyuan: the Association for Computational Linguistics and Chinese Language Processing, 2021: 151-162.
[26] DU X, CARDIE C. Identifying where to focus in reading comprehension for neural question generation[C]//Proceedings of the 2017 conference on empirical methods in natural language processing. Copenhagen: ACL, 2017.
[27] AGRAWAL A, SHUKLA P. Context aware automatic subjective and objective question generation using fast text to text transfer learning[J]. International journal of advanced computer science and applications, 2023, 14(4): 456-463.
[28] FARQUHAR S, KOSSEN J, KUKN L, et al. Detecting hallucinations in large language models using semantic entropy[J]. Nature, 2024, 630(8017): 625-630.
[29] OLNEY A, GRAESSER A, PERSON N. Question generation from concept maps[J]. Dialogue &discourse, 2012, 3(2): 75-99.
[30] YAO L, PENG N, WEISCHEDEL R, et al. Plan-and-write: towards better automatic storytelling[C]//Proceedings of the 33rd AAAI conference on artificial intelligence and 31st innovative applications of artificial intelligence conference and 9th AAAI symposium on educational advances in artificial intelligence. Hawaii: AAAI Press, 2019: 7378-7385.
[31] 陆伟, 李鹏程, 张国标. 学术文本词汇功能识别——基于BERT向量化表示的关键词自动分类研究[J]. 情报学报, 2020, 39(12): 1320-1329. (LU W, LI P C, ZHANG G B. Recognition of lexical functions in academic texts: automaticclassification of keywords based on BERT vectorization[J]. Journal of the China Society for Scientific and Technical Information, 2020, 39(12): 1320-1329.)
[32] 张吉玉, 张均胜. 考虑时序的单篇科技文献新颖性评估方法[J]. 图书情报工作, 2022, 66(17): 93-105. (ZHANG J Y, ZHANG J S. Novelty evaluation method of single scientific and technical literature considering time series[J]. Library and information service, 2022, 66(17): 93-105.)
[33] 朱丽萍. 一种面向科技文献引言的信息抽取方法[J]. 山东大学学报(理学版), 2015, 50(7): 23-30, 37. (ZHU L P. An information extraction method for scientific literature introduction[J]. Journal of Shandong University(natural science), 2015, 50(7): 23-30, 37.)
[34] 刘熠, 张智雄, 王宇飞, 等. 基于语步识别的科技文献结构化自动综合工具构建[J]. 数据分析与知识发现, 2024, 8(2): 65-73. (LIU Y, ZHANG Z X, WANG Y F. Constructing automatic structured synthesis tool for Sci-Tech literature based on move recognition[J]. Data analysis and knowledge discovery, 2024, 8(2): 65-73.)
[35] REIMERS N, GUREVYCH I. Sentence-BERT: sentence embeddings using siamese BERT-networks[C]//Proceedings of the 2019 conference on empirical methods in natural language processing. Hong Kong: ACL, 2019: 296-310.
[36] SHAO Y, GENG Z, LIU Y, et al. CPT: a pre-trained unbalanced transformer for both Chinese language understanding and generation[J]. Science China-information sciences, 2024, 67(5): 152102-1–152102-13.
[37] PAPINENI K, ROUKOS S, WARD T, et al. Bleu: a method for automatic evaluation of machine translation[C]//Proceedings of the 40th annual meeting on Association for Computational Linguistics. Philadelphia: ACL, 2002: 311-318.
[38] LIN C. ROUGE: a package for automatic evaluation of summaries[C]//Proceedings of the workshop on text summarization branches out. Barcelona: ACL, 2004: 74-81.
Options
文章导航

/

〈 〉