图书情报工作

图书情报工作 >

2026 , Vol. 70 >Issue 17: 4 - 20

DOI: https://doi.org/10.13266/j.issn.0252-3116.2026.17.001

专题：大语言模型支持下的文化与典籍计算人文研究

大语言模型强化学习驱动的文化遗迹叙事文本语义组织方法研究

  • 张卫 ,
  • 高鑫 ,
  • 张予歌
展开
  • 1 南京农业大学信息管理学院 南京 210095;
    2 南京农业大学人文与社会计算研究中心 南京 210095
张卫，助理研究员，博士后，E-mail：qwzhang@njau.edu.cn；高鑫，硕士研究生；张予歌，本科生。

收稿日期: 2025-06-20

  修回日期: 2025-10-01

  网络出版日期: 2026-08-20

基金资助

本文系国家自然科学基金青年项目“事件情感知识关联驱动下文化遗迹数字记忆重构模式研究”（项目编号：72404131）和江苏省社会科学基金青年项目“历代经典诗作隐喻的跨语言知识组织及应用研究”（项目编号：24TQC009）研究成果之一。

收起

A Study on the Semantic Organization Method of Cultural Relic Narrative Texts Driven by Large Language Model Reinforcement Learning

  • Zhang Wei ,
  • Gao Xin ,
  • Zhang Yuge
Expand
  • 1 College of Information Management, Nanjing Agricultural University, Nanjing 210095;
    2 Research Center for Humanities and Social Computing, Nanjing Agricultural University, Nanjing 210095
Zhang Wei, Assistant Researcher, Postdoc, E-mail: qwzhang@njau.edu.cn; Gao Xin, Master’s Candidate; Zhang Yuge, Undergraduate Student.

Received date: 2025-06-20

  Revised date: 2025-10-01

  Online published: 2026-08-20

Supported by

This work is supported by the project of National Natural Science Foundation of China, titled “A Study on Digital Memory Reconstruction Mode of Cultural Relics Driven by Event Sentimental Knowledge Association” (Grant No. 72404131); and the Jiangsu Social Science Foundation project titled “A Study of Cross-Language Knowledge Organization and Application of Metaphors in Classic Chinese Poetry of All Ages” (Grant No. 24TQC009).

Fold

摘要

[目的/意义]文化遗迹叙事文本是我国重要的文物数据资源，利用AIGC技术挖掘与组织叙事文本内的事件语义特征，对于文化遗产数字记忆保护与开发具有重要意义。[方法/过程]在大语言模型强化学习驱动下提出基于事件本体的文化遗迹叙事文本语义组织方法。首先，综合专家智慧与文本特征进行事件概念揭示与知识单元重组，归纳遗迹叙事文本中核心元数据形成遗迹事件本体；基于DeepSeek深度思考范式进行事件抽取冷启动以获取伪标签，引入人智协同思维对各任务问答语料进行质量优化；重点针对事件抽取的任务框架设计高效自适应的奖励函数，并通过组内相对优势计算探索事件提及抽取、事件分类、事件论元识别等大模型强化学习的有效性。[结果/结论]经过群体相对策略优化的DeepSeek-R1-Distill-Qwen-32B蒸馏模型分别在事件提及抽取（BLEU＿1=80.21%、ROUGE-1-F1=84.64%）与事件论元识别（F1=78.90%）任务取得最佳实践，强化后的QwQ-32B在事件分类任务上更具优势（ACC=86.09%）；经过实体消歧与共指消解能够进一步优化事件知识图谱的质量，从历史情节分析、人物知识谱系、时空耦合分析方面实现文化遗迹数字叙事服务。

关键词： 数字人文; 文化遗迹; 语义组织; 事件抽取; 大语言模型; 强化学习

本文引用格式

张卫 , 高鑫 , 张予歌 . 大语言模型强化学习驱动的文化遗迹叙事文本语义组织方法研究[J]. 图书情报工作, 2026 , 70(17) : 4 -20 . DOI: 10.13266/j.issn.0252-3116.2026.17.001

Abstract

[Purpose/Significance] Narrative texts of cultural relics are important cultural heritage data resources in China. Using AIGC technology to mine and organize the semantic features of events within these texts is of great significance for the protection and development of the digital memory of cultural heritage. [Method/Process] This paper proposes a semantic organization method for cultural heritage narrative texts based on event ontology, driven by large language model(LLM) reinforcement learning. Firstly, event concepts are revealed and knowledge units are reorganized by integrating expert wisdom with text features, and the core metadata in the relic narrative texts are summarized to form a relic event ontology. Secondly, the DeepSeek reasoning paradigm is utilized for the cold-start of event extraction to obtain pseudo-labels, and human-AI collaboration is introduced to optimize the quality of the corpus. Finally, this paper focuses on designing efficient and adaptive reward functions for the task framework of event extraction, and explores the effectiveness of LLM reinforcement learning for event mention extraction, event classification, and event argument recognition through group relative advantage computation. [Result/Conclusion] The DeepSeek-R1-Distill-Qwen-32B model, optimized with group relative policy optimization, achieves the best performance in the event mention extraction(BLEU＿1=80.21%, ROUGE-1-F1=84.64%) and event argument recognition(F1=78.90%) tasks, respectively. The reinforced QwQ-32B demonstrates superior performance in the event classification task(ACC=86.09%). Entity disambiguation and coreference resolution can further optimize the quality of the event knowledge graph, enabling digital narrative services for cultural relics in terms of historical plot analysis, character knowledge genealogy, and spatiotemporal coupling analysis.

Key words： digital humanities; cultural relics; semantic organization; event extraction; large language model; reinforcement learning

参考文献

[1] 王景慧.从文物保护单位到历史建筑——文物古迹保护方法的深化[J]. 城市规划, 2011, 35(S1):45-47.(WANG J H. From cultural relic protection units to historic buildings:an in-depth study on protection methods of cultural relics[J]. City planning review, 2011, 35(S1):45-47.)
[2] 王晓光,侯西龙,程航航,等.敦煌壁画叙词表构建与关联数据发布[J]. 中国图书馆学报, 2020, 46(4):69-84.(WANG X G, HOU X L, HANG C C, et al. The construction and linked data publishing of Dunhuang mural thesaurus[J]. Journal of library science in China, 2020, 46(4):69-84.)
[3] HU X, NG J, XIA S. User-Centered evaluation of metadata schema for nonmovable cultural heritage:murals and stone cave temples[J]. Journal of the Association for Information Science and Technology, 2018, 69(12):1476-1487.
[4] 张云中,汪诗雨.基于知识图谱的井冈山革命文化资源数字叙事模型的构建与实现[J]. 图书情报工作, 2024, 68(23):106-119.(ZHANG Y Z, WANG S Y. Construction and implementation of a digital narrative model for revolutionary cultural resources in Jinggangshan based on knowledge graph[J]. Library and information service, 2024, 68(23):106-119.)
[5] GUO D, YANG D, ZHANG H, et al. DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning[J]. Nature,2025, 645(8081):633-638.
[6] SCHULMAN J, WOLSKI F, DHARIWAL P, et al. Proximal policy optimization algorithms[PP]. arXiv preprint arXiv:1707.06347, 2017.
[7] RAFAIOV R, SHARMA A, MITCHELL E, et al. Direct preference optimization:your language model is secretly a reward model[J]. Advances in neural information processing systems, 2023, 36:53728-53741.
[8] ZHAN Z, ZHOU S, ZHOU H, et al. An evaluation of DeepSeek models in biomedical natural language processing[PP]. arXiv preprint arXiv:2503.00624, 2025.
[9] 高劲松,韩牧哲.考古发掘资料图数据库的语义关联构建研究[J]. 图书情报工作, 2021, 65(9):105-116.(GAO J S, HAN M Z. Research on semantic association construction of the graph database on archaeological excavation resources[J]. Library and information service, 2021, 65(9):105-116.)
[10] 范涛,王昊,陈玥彤.基于深度迁移学习的地方志多模态命名实体识别研究[J]. 情报学报, 2022, 41(4):412-423.(FAN T, WANG H, CHEN Y T. Research on multimodal named entity recognition of local history based on deep transfer learning[J]. Journal of the China Society for Scientific and Technical Information, 2022, 41(4):412-423.)
[11] FAN Z, CHEN C. CuPe-KG:Cultural perspective–based knowledge graph construction of tourism resources via pretrained language models[J]. Information processing&management,2024, 61(3):103646.
[12] KEATS P A. Multiple text analysis in narrative research:Visual,written, and spoken stories of experience[J]. Qualitative research,2009, 9(2):181-195.
[13] 宋宁远,王晓光.基于情节本体的叙事性文本语义结构化表示方法研究[J]. 中国图书馆学报, 2020, 46(2):96-113.(SONG N Y, WANG X G. Semantic structured representation method of narrative text based on plot ontology[J]. Journal of library science in China, 2020, 46(2):96-113.)
[14] 刘宗田,黄美丽,周文,等.面向事件的本体研究[J]. 计算机科学, 2009, 36(11):189-192.(LIU Z T, HUANG M L, ZHOU W,et al. Research on event-oriented ontology model[J]. Computer science, 2009, 36(11):189-192.)
[15] KOWALSKI R, SERGOT M. A logic-based calculus of events[J]. New generation computing, 1986, 4:67-95.
[16] GUNAL S, GEREK O N, ECE D G, et al. The search for optimal feature set in power quality event classification[J]. Expert systems with applications, 2009, 36(7):10266-10273.
[17] JEONG S, KIM H G. SEDE:an ontology for scholarly event description[J]. Journal of information science, 2010, 36(2):209-227.
[18] DOERR M. The CIDOC conceptual reference module:an ontological approach to semantic interoperability of metadata[J]. AI magazine, 2003, 24(3):75-75.
[19] GUAN S, CHENG X, BAI L, et al. What is event knowledge graph:A survey[J]. IEEE Transactions on knowledge and data engineering, 2023, 35(7):7569-7589.
[20] LI Q, LI J, SHENG J, et al. A survey on deep learning event extraction:approaches and applications[J]. IEEE transactions on neural networks and learning systems, 2022.
[21] YANG S, FENG D, QIAO L, et al. Exploring pre-trained language models for event extraction and generation[C] //Proceedings of the 57th annual meeting of the Association for Computational Linguistics. 2019:5284-5294.
[22] LU Y, LIN H, XU J, et al. Text2Event:controllable sequenceto-structure generation for end-to-end event extraction[C] //Proceedings of the 59th annual meeting of the Association for Computational Linguistics and the 11th international joint conference on natural language processing(volume 1:long papers). 2021:2795-2806.
[23] CHEN R, QIN C, JIANG W, et al. Is a large language model a good annotator for event extraction?[C] //Proceedings of the AAAI conference on artificial intelligence. 2024, 38(16):17772-17780.
[24] REN Y, CAO Y, GUO P, et al. Retrieve-and-sample:document-level event argument extraction via hybrid retrieval augmentation[C] //Proceedings of the 61st annual meeting of the Association for Computational Linguistics(volume 1:long papers). 2023:293-306.
[25] HUANG F, HUANG Q, ZHAO Y, et al. A three-stage framework for event-event relation extraction with large language model[C] //International conference on neural information processing.Singapore:Springer, 2023:434-446.
[26] YAN Y, LIU J, JI D, et al. Collaborate SLM and LLM with latent answers for event detection[J]. Knowledge-based systems, 2024,305:112684.
[27] SRIVASTAVA S, YAO Z. Revisiting prompt optimization with large reasoning models:a case study on event extraction[PP]. arXiv preprint arXiv:2504.07357, 2025.
[28] 司莉,周璟.面向保护和利用的大遗址本体与关联数据研究[J]. 图书馆论坛, 2024, 44(1):105-115.(SI L, ZHOU J. On the protection and utilization of great sites based on ontology and linked data technologies[J]. Library tribune, 2024, 44(1):105-115.)
[29] ACIERNO M, CURSI S, SIMEONE D, et al. Architectural heritage knowledge modelling:an ontology-based framework for conservation process[J]. Journal of cultural heritage, 2017, 24:124-133.
[30] LI M, ZHOU H, YANG H, et al. RT:a retrieving and chainof-thought framework for few-shot medical named entity recognition[J]. Journal of the American Medical Informatics Association, 2024, 31(9):1929-1938.
[31] WANG D, CHURCHILL E, MAES P, et al. From human-human collaboration to human-AI collaboration:designing AI systems that can work together with people[C] //Extended abstracts of the2020 CHI conference on human factors in computing systems.2020:1-6.
[32] SHAO Z, WANG P, ZHU Q, et al. DeepSeekmath:Pushing the limits of mathematical reasoning in open language models[PP]. arXiv preprint arXiv:2402.03300, 2024.
[33] PAPINENI K, ROUKOS S, WARD T, et al. Bleu:a method for automatic evaluation of machine translation[C] //Proceedings of the 40th annual meeting of the Association for Computational Linguistics. 2002:311-318.
[34] LIN C Y. Rouge:a package for automatic evaluation of summaries[C] //Text summarization branches out. 2004:74-81.
Options
文章导航

/

〈 〉