面向智能楼宇运维知识图谱构建的事件联合抽取方法

Event Joint Extraction Method for the Construction of Intelligent Building Operation and Maintenance Knowledge Graphs

摘要

楼宇作为电力系统最末端用能单元,在突发运维事件发生时,高效的故障诊断与检修工作直接关乎供电连续性与用户用电体验．知识图谱能够高效抽取楼宇运行过程中的非结构化数据实体及其关联属性,为故障诊断与智能运维提供有力支持．但传统抽取方法在多触发词与论元关联多事件的场景下抽取完整率低,难以有效支撑楼宇运维知识图谱构建．对此,提出一种面向楼宇运维知识图谱构建的事件联合抽取方法．首先,研究基于点互信息与左右熵的自适应阈值分词算法,精确切分运维领域文本字词；其次,定义楼宇运维知识图谱多元关系统一表达范式,应对同一论元具备多重角色的问题；然后,训练领域专属的大语言模型——LouYSpanBERT,构建触发词边界标记驱动的楼宇运维事件联合抽取框架,通过“触发词检测-边界标记-论元抽取”三级联合抽取提高事件抽取精度；最后,以某园区运维数据做实例分析,与4个已有模型比较,所提方法执行典型的论元分类AC任务时,F1值有效提升了5.08%、9.81%、13.36%、14.99%,并且本文将抽取结果通过Neo4j实现楼宇运维知识图谱可视化展示．

Abstract

As the final energy-consuming unit of a power system,buildings directly affect the continuity of power supply and the user’s electricity experience. Therefore,efficient fault diagnosis and repair are crucial in the event of sudden operational breakdowns and maintenance events. Knowledge graphs can efficiently extract unstructured data entities and their associated attributes during construction,thereby providing strong support for fault diagnosis and intelligent operation and maintenance. However,traditional extraction methods achieve low extraction completeness in scenarios with multiple trigger words and multiple events associated with argument elements,making it difficult to effectively support the construction of building operation and maintenance knowledge graphs. In this regard,we develop an event joint extraction method for graph construction. First,an adaptive threshold word segmentation algorithm based on point mutual information and left-right entropy is studied to accurately segment text words and characters in the operation and maintenance field. Second,the unified expression paradigm of multiple relationships in the building operation and maintenance knowledge graph is designed to address the issue of the same argument having multiple roles. Next,a domain-specific large language model,LouYSpanBERT,is trained,and a joint extraction framework is constructed to extract operation and maintenance events driven by trigger word boundary markers. The accuracy of event extraction is improved through a three-level joint extraction of “trigger word detection-boundary marker-argument extraction.” Finally,the operation and maintenance data for a specific park are used as an example. Compared with four existing models,when the proposed method performs the typical argument classification(AC) task,the F1-score effectively increases by 5.08%,9.81%,13.36%,and 14.99%,respectively. Here,we visually display the extraction results in Neo4j to build an operational and maintenance knowledge graph.

关键词

事件联合抽取 / 知识图谱 / 分词算法 / 触发词边界 / 论元分类

Key words

event joint extraction / knowledge graph / word segmentation algorithm / trigger word boundary / argument classification

参考文献

[1] 张新长, 华淑贞, 齐霁, 等. 新型智慧城市建设与展望:基于AI的大数据、大模型与大算力[J]. 地球信息科学学报, 2024, 26(4):779-789.

[2] Zhang Xinchang, Hua Shuzhen, Qi Ji, et al. Progress and prospects of new smart city construction:AI—based big data,big models and big computing power[J]. Journal of Geo—Information Science, 2024, 26(4):779-789(in Chinese).

[3] Dai Y X, Hasanefendic S, Bossink B. A systematic literature review of the smart city transformation process:The role and interaction of stake—holders and technology[J]. Sustainable Cities and Society, 2024, 101:1-29.

[4] Zhong L F, Wu J, Li Q, et al. A comprehensive survey on automatic knowledge graph construction[J]. ACM Computing Surveys, 2023, 56(4):1-62.

[5] 闫玮丹, 齐冬莲, 闫云凤, 等. 面向电力领域的知识图谱与大模型融合关键技术及其典型应用[J]. 高电压技术, 2025, 51(4):1747-1762.

[6] Yan Weidan, Qi Donglian, Yan Yunfeng, et al. Key technologies and typical applications of knowledge graph and large language model fusion in the power sector[J]. High Voltage Engineering, 2025, 51(4):1747-1762(in Chinese).

[7] 张禹方, 袁之康, 高硕杰, 等. 基于跨模态数据的变压器套管故障知识图谱构建与应用[J]. 中国电机工程学报, 2025, 45(22):9064-9074.

[8] Zhang Yufang, Yuan Zhikang, Gao Shuojie, et al. Construction and application of knowledge graph of transformer bushing faults based on cross—modal data[J]. Proceedings of the CSEE, 2025, 45(22):9064-9074(in Chinese).

[9] Yamashita D Y, Vechiu I, Gaubert J P. A review of hierarchical control for building microgrids[J]. Renewable and Sustainable Energy Reviews, 2020, 118:109523.

[10] 束嘉伟, 杨挺, 耿毅男, 等. 面向电力知识图谱构建的重叠实体关系联合抽取方法[J]. 高电压技术, 2024, 50(11):4912-4922.

[11] Shu Jiawei, Yang Ting, Geng Yinan, et al. Joint extraction method for overlapping entity relationships in the construction of electric power knowledge graph[J]. High Voltage Engineering, 2024, 50(11):4912-4922(in Chinese).

[12] Xu J H, Yang C, Kang X J. LAAP:Learning the argument of an entity with event prompts for document—level event extraction[J]. Neurocomputing, 2025, 613:128-139.

[13] Li Q, Li J X, Sheng J W, et al. A survey on deep learning event extraction:Approaches and applications[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 35(5):6301-6321.

[14] Wei Y, Bo L L, Sun X B, et al. Automated event extraction of CVE descriptions[J]. Information and Software Technology, 2023, 158:107-118.

[15] Guan Y, Chen J Y, Lecue F, et al. Trigger—argument based explanation for event detection[C]// Findings of the Association for Computational Linguistics:ACL 2023. Toronto,Canada, 2023:5046-5058.

[16] Yang J, Han S C, Poon J. A survey on extraction of causal relations from natural language text[J]. Knowledge and Information Systems, 2022, 64(5):1161-1186.

[17] Peng J R, Yang W Z, Wei F Y, et al. Prompt for extraction:Multiple templates choice model for event extraction[J]. Knowledge—Based Systems, 2024, 289:111-124.

[18] 刘万里, 雍新有, 曹开臣, 等. 基于提示学习的ERNIE—BiLSTM—PN通用信息抽取方法研究[J]. 电子科技大学学报, 2025, 54(3):411-423.

[19] Liu Wanli, Yong Xinyou, Cao Kaichen, et al. Universal information extraction method based on prompt learning with ERNIE—BiLSTM—PN[J]. Journal of University of Electronic Science and Technology of China, 2025, 54(3):411-423(in Chinese).

[20] 齐云雷, 王冰. 基于span指针网络的重叠和嵌套事件抽取联合学习框架[J/OL]. 控制工程, https://doi.org/10.14107/j.cnki.kzgc.20240569.

[21] Qi Yunlei, Wang Bing. A framework for joint learning of overlapping and nested event extraction based on span pointer networks[J/OL]. Control Engineering of China, https://doi.org/10.14107/j.cnki.kzgc.20240569(in Chinese).

[22] Padmakumar V, He H. Unsupervised Extractive Summarization Using Pointwise Mutual Information[EB/OL]. https://arxiv.org/abs/2102.ob272, 2021—02—11.

[23] Li Q, Guo S, Wu J, et al. Event extraction by associating event types and argument roles[J]. IEEE Transactions on Big Data, 2023, 9(6):1549-1560.

[24] Gardazi N M, Daud A, Malik M K, et al. BERT applications in natural language processing:A review[J]. Artificial Intelligence Review, 2025, 58(166):1-49.

[25] Joshi M, Chen D Q, Liu Y H, et al. SpanBERT:Improving pre—training by representing and predicting spans[J]. Transactions of the Association for Computational Linguistics, 2020, 8:64-77.

[26] Zhu Z J, Dai W H, Hu Y, et al. Speech emotion recognition model based on Bi—GRU and focal loss[J]. Pattern Recognition Letters, 2020, 140:358-365.

[27] An Y, Xia X Y, Chen X L, et al. Chinese clinical named entity recognition via multi—head self—attention based BiLSTM—CRF[J]. Artificial Intelligence in Medicine, 2022, 127:102-112.

[28] Veličković P, Buesing L, Overlan M C, et al. Pointer graph networks[J]. Advances in Neural Information Processing Systems, 2020, 33:2232-2244.

[29] Yao H S, Zhu D L, Jiang B, et al. Negative log likelihood ratio loss for deep neural network classification[C]// Proceedings of the Future Technologies Conference(FTC). San Francisco,USA, 2019:276-282.

[30] Sheng J W, Guo S, Yu B W, et al. CasEE:A joint learning framework with cascade decoding for overlapping event extraction[C]// Findings of the Association for Computational Linguistics:ACL—IJCNLP 2021. 2021:164-174.

[31] Yang S, Feng D W, Qiao L B, 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. Florence,Italy, 2019:5284-5294.

[32] Du X Y, Cardie C. Document—level event role filler extraction using multi—granularity contextualized encoding[C]// Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. 2020:8010-8020.

[33] Cao H, Li J Y, Su F F, et al. OneEE:A one—stage framework for fast overlapping and nested event extraction[C]// Proceedings of the 29th International Conference on Computational Linguistics. Gyeongju,Republic of Korea, 2022:1953-1964.

基金资助

智能电网国家科技重大专项(2030)资助项目(2024ZD0800800)

国家自然科学基金资助项目(62371338)