Urban Waterlogging Information Recognition Method Based on MacBERT and Adversarial Training
Author:
Affiliation:

Funding:

National Key Research and Development Program of China (Urban Management and Service Technology for Physical and Digital Spatial Integration: 2019YFB2102503), Basic Research Funds of the Chinese Academy of Surveying and Mapping (Key Technology Research of Geospatial Big Data Governance: AR2111)

Ethical statement:

  • Article
  • |
  • Figures
  • |
  • Metrics
  • |
  • Reference
  • |
  • Related
  • |
  • Cited by
  • |
  • Materials
    Abstract:

    Methods such as BERT and the combination of neural network model have been gradually applied to the acquisition of disaster information. However, such methods have many problems, such as large number of parameters, inconsistent data sets and fine-tuning data sets, and local instability. In this paper, an information recognition model based on MacBERT and adversarial training is proposed. The model obtains the initial vector representation through MacBERT pre-training model, and then adds some perturbations to generate adversarial samples. Then input to the bi-directional long short-term memory and conditional random field in turn, which not only reduces the pre-training times and fine-tuning stage differences, but also improves the robustness of the model. The experimental results show that the information recognition model based on MacBERT and adversarial training are improved the accuracy rate and F1 value on the microblog dataset and the 1998 People’s Daily dataset, and the execution is excellent than other models, which indicates that the model has certain feasibility for urban waterlogging information recognition.

    Reference
    Related
    Cited by
Get Citation

FANG Meili, ZHENG Yingying, TAO Kunwang, ZHAO Xizhi, QIU Agen, LU Wen. Urban Waterlogging Information Recognition Method Based on MacBERT and Adversarial Training[J]. Journal of Integration Technology,2023,12(1):56-67

Copy
Share
Article Metrics
  • Abstract:
  • PDF:
  • HTML:
History
  • Received:
  • Revised:
  • Adopted:
  • Online: January 12,2023
  • Published: