1.School of Computer Science and Technology,North University of China,Taiyuan 030051,China
2.Shaanxi Aircraft Industry Co. ,Ltd. ,Hanzhong 723200,China
Show less
文章历史+
Received
Published
2024-11-21
2025-10-31
Issue Date
2025-11-27
PDF (1708K)
摘要
专利文本长度较长且关系重叠严重, 传统的关系抽取模型中将关系视作离散的标签且只关注文本的局部特征, 未充分利用其中的潜在信息, 影响复杂的专利文本的抽取性能。针对以上问题, 本文在OneRel模型的基础上提出了一种基于关系标签语义与全局特征融合的填表式实体关系抽取模型SRL-GFI(Semantics of Relation Labels and Global Feature Integration)。SRL-GFI模型使用BERT预训练模型获取关系标签的语义信息, 并与表格标记方式相结合得到关系-标记联合编码, 使之与分词对的编码进行匹配得出相应的相关性分数, 再通过全局信息融合模块将模型中间结果反馈回去并迭代, 以融合其中的全局特征, 从而进行实体关系抽取。多个数据集上的对比实验都显示了新模型性能的提升, 其中在专利实体抽取数据集PERD和TFH2020上, 1值分别达到79.8%和55.9%, 相比OneRel提升了5.0百分点和1.8百分点。本文提出的SRL-GFI模型有效利用了关系标签中的语义信息, 以及文本中的全局信息和关系三元组中的内在联系, 可以很好地应用于专利领域中复杂文本的实体关系抽取。
Abstract
Patent texts are lengthy and contain heavily overlapping relationships, while traditional relation extraction models treat relations as discrete labels and focus only on local features of the text, failing to fully utilize the latent information, thereby affecting the extraction performance on complex patent texts. To address these issues, this paper proposed a table-filling entity relation extraction model SRL-GFI (Semantics of Relation Labels and Global Feature Integration), based on the OneRel model, which integrated the semantics of relation labels and global features. The SRL-GFI model utilized a BERT pre-trained model to obtain semantic information of relation labels and combined it with tagging scheme to derive a relation-marking joint encoding. This encoding was then matched with embedding of word pairs to obtain the corresponding relevance scores. Subsequently, the global feature integrating module fed back and iterated the intermediate results of the model to integrate global features, thereby performing entity relation extraction. Comparative experiments on multiple datasets show performance improvements of the new model, with scores reaching 79.8% and 55.9% on the patent entity extraction datasets PERD and TFH2020, respectively, representing improvements of 5.0 and 1.8 percentage points over OneRel. The SRL-GFI model proposed in this paper effectively utilizes the semantic information in relation labels, as well as global information in the text and intrinsic connections within relation triplets, making it well-suited for entity relation extraction of complex texts in the patent domain.
CHENLiang, CHENLili, XUHaiyun, et al. A global literature review in recent advancement of patent mining[J]. Library and Information Service, 2024, 68(2): 110-133. (in Chinese)
[3]
GOLSHANP N, DASHTIH R, AZIZIS, et al. A study of recent contributions on information extraction [DB/OL]. (2018-05-15)[2024-11-21].
LIChengqi, LEIHaiwei, LIFan, et al. Method for extracting patent entity relationships based on nearest pairing addressing[J]. Computer Engineering and Design, 2024, 45(4): 1100-1108. (in Chinese)
[8]
SHANGY M, HUANGH, MAOX. Onerel: Joint entity and relation extraction with one module in one step[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2022, 36(10): 11285-11293.
HaihongE, ZHANGWenjing, XIAOSiqi, et al. Survey of entity relationship extraction based on deep learning[J]. Journal of Software, 2019, 30(6): 1793-1818. (in Chinese)
[11]
CHANY S, ROTHD. Exploiting syntactico-semantic structures for relation extraction [C]//49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, 2011: 551-560.
[12]
GORMLEYM R, YUM, DREDZEM. Improved relation extraction with feature-rich compositional embedding models[C]//2015 Conference on Empirical Methods in Natural Language Processing, 2015: 1774-1784.
[13]
LIQ, JIH. Incremental joint extraction of entity mentions and relations[C]//52nd Annual Meeting of the Association for Computational Linguistics, 2014: 402-412.
[14]
WEIZ, SUJ, WANGY, et al. A novel cascade binary tagging framework for relational triple extraction[C]//58th Annual Meeting of the Association for Computational Linguistics, 2020: 1476-1488.
[15]
YANZ, ZHANGC, FUJ, et al. A partition filter network for joint entity and relation extraction[C]//2021 Conference on Empirical Methods in Natural Language Processing, 2021: 185-197.
[16]
ZHENGH, WENR, CHENX, et al. PRGC: Potential relation and global correspondence based joint relational triple extraction[C]//59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, 2021: 6225-6235.
[17]
WANGY, YUB, ZHANGY, et al. TPLinker: Single-stage joint extraction of entities and relations through token pair linking[C]//28th International Conference on Computational Linguistics, 2020: 1572-1582.
[18]
RENF, ZHANGL, YINS, et al. A novel global feature-oriented relational triple extraction model based on table filling[C]//2021 Conference on Empirical Methods in Natural Language Processing, 2021: 2646-2656.
[19]
WANGY, SUNC, WUY, et al. UniRE: A unified label space for entity relation extraction[C]//59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, 2021: 220-231.
[20]
ZHANGJ, JIANGX, SUNY, et al. RS-TTS: A novel joint entity and relation extraction model[C]//26th International Conference on Computer Supported Cooperative Work in Design (CSCWD), 2023: 71-76.
[21]
NINGJ, YANGZ, SUNY, et al. OD-RTE: A one-stage object detection framework for relational triple extraction[C]//61st Annual Meeting of the Association for Computational Linguistics, 2023: 11120-11135.
[22]
DAIQ, YANGW, WANGL, et al. SOIRP: Subject-object interaction and reasoning path based joint relational triple extraction by table filling[J]. Neurocomputing, 2024, 580: 127492.
[23]
XUB, WANGQ, LYU Y, et al. EmRel: Joint representation of entities and embedded relations for multi-triple extraction[C]//2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2022: 659-665.
[24]
CHENGJ, ZHANGT, ZHANGS, et al. A cascade dual-decoder model for joint entity and relation extraction [J]. IEEE Transactions on Emerging Topics in Computational Intelligence, 2025, 9: 1130-1142.
[25]
TANGW, XUB, ZHAOY, et al. UniRel: Unified representation and interaction for joint relational triple extraction[C]//2022 Conference on Empirical Methods in Natural Language Processing, 2022: 7087-7099.
[26]
CHENL, XUS, ZHUL, et al. A deep learning based method for extracting semantic information from patent documents[J]. Scientometrics, 2020, 125(1): 289-312.
DENGNa, YUZhuoqun, DANWenjun, et al. A joint extraction model for entity relationship in traditional Chinese medicine patent texts based on semantic features and multi-layer cross-attention mechanism[J].Data Analysis and Knowledge Discovery, 2025, 9(7): 141-153. (in Chinese)
HEYu, ZHANGXiaodong, ZHENGXin. Constructing patent knowledge graph with spert-aggcn model[J].Data Analysis and Knowledge Discovery, 2024, 8(1): 146-156. (in Chinese)
[31]
王腾科.面向中文专利文本的命名实体识别及关系抽取研究[D].淮南: 安徽理工大学, 2024.
[32]
VASWANIA, SHAZEERN, PARMARN . et al. Attention is all you need[DB/OL]. (2017-06-12) [2024-11-21].
[33]
RIEDELS, YAOL, MCCALLUMA. Modeling relations and their mentions without labeled text[C]//European Conference on Machine Learning and Knowledge Discovery in Databases(ECML PKDD 2010), 2010: 148-163.
[34]
ROTHD, YIH W. A linear programming formulation for global inference in natural language tasks[C]//8th Conference on Computational Natural Language Learning (CoNLL-2004) at HLT-NAACL 2004, 2004: 1-8.
[35]
LUANY, HEL, OSTENDORFM, et al. Multi-task identification of entities, relations, and coreference for scientific knowledge graph construction[C]//2018 Conference on Empirical Methods in Natural Language Processing, 2018: 3219-3232.
[36]
ZHAOK, XUH, CHENGY, et al. Representation iterative fusion based on heterogeneous graph neural network for joint entity and relation extraction[J]. Knowledge-Based Systems, 2021, 219: 106888.