多维特征学习与模型融合的地理命名实体识别

doi:10.13474/j.cnki.11-2246.2024.0923

摘要/Abstract

摘要： 地理命名实体识别是地理信息抽取的核心任务,而地理信息抽取支撑着地理信息系统的构建。但目前的地理命名实体识别研究面临两大核心挑战:一是地理领域文本的标注数据稀缺,导致传统依赖大量标注数据的通用模型难以全面捕捉并识别地理文本中所有潜在的命名实体;二是地理数据的标签密度较为稀疏,模型在区分不同地理命名实体时往往无法区分其边界,进而无法精准定位。针对上述问题,本文提出了一种面向地理领域文本特征的命名实体识别算法AM_NER。首先,利用Albert进行词向量训练,该模型是面向小样本的轻量级预训练模型,能够更为全面地学习地理领域的语义信息;然后,设计了名为M_NER的神经元结构,该神经元基于模型融合思想,利用多个模型从不同维度对语义特征进行学习,进而准确识别出命名实体的边界。相较于此前的研究,AM_NER在地理领域数据集中的各项指标提升了2.05%～2.67%。

关键词: 地理命名实体识别, 深度学习, 特征学习, 模型融合

Abstract: Geographical named entity recognition is the core task of geographic information extraction, which supports the construction of geographic information systems. However, current research on geographic named entity recognition faces two core challenges: Firstly, the scarcity of annotated data in geographic texts makes it difficult for traditional generic models that rely heavily on annotated data to fully capture and recognize all potential named entities in geographic texts.Secondly, the label density of geographic data is relatively sparse, and models often can not distinguish the boundaries of different geographic named entities, thus unable to accurately locate them. In response to the above issues, this study proposes a named entity recognition algorithm AM-NER for geographic text features. Firstly, using Albert for word vector training, this model is a lightweight pre training model for small samples, which can comprehensively learn semantic information in the geographic field.Secondly, a neuron structure named MNER is designed, which is based on the idea of model fusion and utilizes multiple models to learn semantic features from different dimensions, accurately identifying the boundaries of named entities. Compared to previous studies, AM-NER has improved various indicators in the geographic dataset by 2.05%～2.67%.

Key words: geographical named entity recognition, deep learning, feature learning, model fusion

中图分类号:

P208

马浩然, 王金华. 多维特征学习与模型融合的地理命名实体识别[J]. 测绘通报, 2024, 0(9): 129-134.

MA Haoran, WANG Jinhua. Geographic named entity recognition based on multi-dimensional feature learning and model fusion[J]. Bulletin of Surveying and Mapping, 2024, 0(9): 129-134.

参考文献

[1] WANG Chengshan, HAZEN R M, CHENG Qiuming, et al.The deep-time digital earth program: data-driven discovery in geosciences[J].National Science Review, 2021, 8(9): nwab027.
[2] 马凯, 田苗, 谭永健, 等.基于四份区域地质调查报告构建的命名实体识别试验数据集研发[J].全球变化数据学报(中英文), 2022(1): 78-84.
[3] 储德平, 万波, 李红, 等.基于ELMO-CNN-BiLSTM-CRF模型的地质实体识别[J].地球科学, 2021(8): 3039-3048.
[4] LAN Z Z, CHEN M, GOODMAN S, et al.Albert: a lite bert for self-supervised learning of language representations[J].Computer Science, 2019: 202888986.
[5] ALAPARTHI S, MISHRA M.Bidirectional encoder representations from transformers (BERT): a sentiment analysis odyssey[J].Computer Science, 2020: 220302509.
[6] HOCHREITER S, SCHMIDHUBER J.Long short-term memory[J].Neural Computation, 1997, 9(8): 1735-1780.
[7] CHUNG J, JUN Y.Empirical evaluation of gated recurrent neural networks on sequence modeling[J].Computer Science, 2014:5201925.
[8] MA Haoran, DING Zhaoyun.Research on multi-model fusion for named entity recognition based on loop parameter sharing transfer learning[C]//Proceedings of 2022 IEEE International Conference on Data Science in Cyberspace (DSC).Guilin: IEEE, 2022.
[9] LI Jing, SHANG Shuo, SHAO Ling.MetaNER: named entity recognition with meta-learning[J].Computer Science, 2020:215875823.
[10] DA I L, LI Y, JIANZHONG Y, et al.Dependency syntax guided BERT-BiLSTM-GAM-CRF for Chinese NER[J].Computer Science, 2022:246963066.
[11] ZHU Peng, CHENG Dawei, YANG Fangzhou, et al.ZH-NER: Chinese named entity recognition with adversarial multi-task learning and self-attentions[M]//Database Systems for Advanced Applications.Cham: Springer International Publishing, 2021: 603-611.
[12] 刘昕, 徐洪珍, 刘爱华, 等.基于MacBERT和R-Drop的地质命名实体识别[J].郑州大学学报(工学版), 2024, 45(3): 89-95.
[13] 王刘坤, 李功权.基于GeoERNIE-BiLSTM-Attention-CRF模型的地理命名实体识别[J].地理科学, 2023, 58(7):1164-1177.
[14] 谢雪景, 谢忠, 马凯, 等.结合BERT与BiGRU-Attention-CRF模型的地质命名实体识别[J].地质通报, 2023, 42(5): 846-855.
[15] LAFFERTY J, MCCALLUM A, PEREIRA F.Conditional random fields: probabilistic models for segmenting and labeling sequence data[J].Computer Science, 2001:219683473.
[16] 姜松言, 宋国民, 贾奋励, 等.结合地名知识和位置信息的中文地名匹配方法[J].测绘通报, 2022(5): 120-125.
[17] MA Haoran, DING Zhaoyun, ZHOU Dongsheng, et al.Research on NER based on register migration and multi-task learning[M]//Wireless Algorithms, Systems, and Applications.Cham: Springer Nature Switzerland, 2022: 657-666.
[18] 杨盈, 邱芹军, 谢忠, 等.人在回路学习增强的地理命名实体识别[J].测绘通报, 2023(8): 155-160.