1.College of Computer Science and Technology,Changchun University,Changchun 130022,China
2.Ministry of Education Key Laboratory of Intelligent Rehabilitation and Barrier-free Access for the Disabled,Changchun University,Changchun 130022,China
3.College of Special Education,Changchun University,Changchun 130022,China
A end-to-end entity relationship extraction model based on span and semantic features was proposed to address the problems of traditional entity relationship extraction methods relying on distance measurement or simple span recognition, which make it difficult to capture potential relationships between entities, and the high computational complexity and error propagation of the model. Firstly, the text vectors were randomly segmented into span sequences so that the model can learn a wider range of semantic feature information. Secondly, semantic relationships were judged to screen out subsets of candidate relationships, thus reducing information redundancy. Finally, the candidate relationships were transformed into relationship-span combinations containing important relationship semantics, and the Transformer decoder was used to achieve the joint extraction of entity relationships. The experimental results show that the F1 value of this model is significantly improved in the NYT and WebNLG datasets compared to other baseline models, proving its effectiveness.
在文本嵌入模块,首先将文本输入来自变换器的双向编码器表示(Bidirectional encoder repre sentations from transformers,BERT)[13]嵌入层,使用BERT预训练语言模型将文本转换为向量,并将向量随机分割成跨度序列。在语义关系判定模块中,使用多头注意力机制筛选最相关的关系子集,过滤不相关的关系,提高关系细粒度识别的精确度,以便对跨度和关系进行联合判定。在实体关系联合抽取模块中,将候选关系转换为包含重要关系语义的关系-跨度组合,并使用Transformer解码器完成三元组抽取任务。
1.2 文本嵌入模块
近年来,以BERT为代表的预训练模型在自然语言处理(Natural language processing,NLP)的各项任务中取得了较先进的效果,也被成功应用到实体识别、关系抽取任务中,因此本文选用BERT作为文本嵌入工具。定义输入文本为,xn 表示文本中的字符,n表示文本序列的长度。将字符级分割后的文本输入BERT模型。BERT预训练模型包含12个隐藏层,每个隐藏层的大小为768。每个字符被映射为对应的向量表示,如下所示:
传统的实体关系抽取一般基于距离度量或简单跨度识别,未能充分挖掘实体之间的潜在关系。例如“Xiao Ming's father is Wang Jian-jun.”中存在“father”这一关系,由于语义的多样性还会出现例如“Wang Jian-jun's son is named Xiao Ming.”的情况,这类隐含关系在文本语义层面十分明显。
LiDong-mei, ZhangYang, LiDong-yuan, et al. Review of studies on entity relationship extraction methods[J]. Computer Research and Development, 2020,57(7): 1424-1448.
[3]
YangY, WuZ L, YangY X, et al.A Survey of information extraction based on deep learning[J].Applied Sciences, 2022, 12(19): No.9691.
[4]
LiJ, SunA X, HanJ L, et al. A survey on deep learning for named entity recognition[J]. IEEE Transactionson Knowledge and Data Engineering, 2020: 50-70.
WangJin, JiangShi-qi. Nested named entity recognition method based on multiplexed local feature integration[J]. Journal of Jiangsu University (Natural Science Edition), 2025, 46(4): 431-437.
WangJin, WangMeng-qi, ZhangXin-yue, et al. Chinese named entity recognition based on multi-head attention character word integration[J]. Journal of Jiangsu University (Natural Science Edition), 2024, 45(1): 77-84.
[9]
TuoM, YangW Z, WeiF Y, et al.A novel chinese overlapping entity relation extraction model using word-label based on cascade binary tagging[J]. Electronics, 2023, 12(4): No.1013.
[10]
WangP. A survey of research on deep learning entity relationship extraction[J]. Natural Language Processing and Speech Recognition, 2019, 1(1): 1-5.
[11]
LiJ L, XuY J, LinH Z, et al.Semantic-consistent learning for one-shot joint entity and relation extraction[J]. Applied Intelligence, 2022,53(5): 5963-5976.
[12]
LinY, ShenS, LiuZ, et al. Neural relation extraction with selective attention over instances[C]∥Proceedings of the 54the Annual Meeting of the Association for Computational Linguistics. Berlin: Association for Computational Linguistics, 2016: 2124-2133.
[13]
KalpitD, YaserA.Span-Level Model for Relation Extraction[C]∥Proceedings of 57th Annual Meeting of the Association for Computational Linguistics (ACL 2019), Florence, Italy, 2019: 5308-5314.
[14]
EbertsM, UlgesA.Span-based joint entity and relation extraction with transformer pre-Training[C]∥ Proceedings of the 24th European Conference on Artificial Intelligence. Amsterdam: IOS Press, 2020: 2006-2013.
[15]
WanQ, WeiL, ZhaoS, et al.A span-based multi-modal attention network for joint entity-relation extraction[J].Knowledge-Based Systems, 2023, 262:No.110228.
[16]
DevlinJ, ChangM W, LeeK, et al. BERT: pre-training of deep bidirectional transformers for language understanding[EB/OL]. [2024-03-15].
[17]
MaL, RenH, ZhangX.Effective cascade dual-decoder model for joint entity and relation extraction[EB/OL]. [2024-03-17].
[18]
SurdeanuM, TibshiraniJ, NallapatiR, ManningC D. Multi-instance multi-label learning for relation extraction[C]∥Proceedings of the Joint Conference on Empirical Methods in Natural Language Processing and Computationl Natural Language Learning. Je ju Island: Association for Computational Linguistics, 2012: 455-465.
[19]
RiedelS, YaoL, McCallumA.Modeling relations and their mentions without labeled text[C]∥Proceedings of the European Conference on Machine Learning and Knowledge Discovery in Databases (ECML PKDD). Berlin: Springer, 2010: 148-163.
[20]
ZengX R, ZengD J, HeS Z,et al.Extracting relational facts by an end-to-end neural model with copy mechanism[C]∥Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics,Melbourne, Australia, 2018: 506-514.
[21]
RenF L, ZhangL H, YinS J,et al.A novel global feature-oriented relational triple extraction model based on table filling[C]∥Proceedings of the Conference on Empirical Methods in Natural Language Processing, Online and Punta Cana, Dominican Republic, 2021: 2646-2656.
[22]
XieX Y, XieM, MoshayediA J, et al. A hybrid improved neural networks algorithm based on L2 and dropout regularization[J].Mathematical Problems in Engineering, 2022: 1-19.