Deep learning-based methods for natural hazard named entity recognition.

Area of Science:

• Natural Language Processing
• Disaster Management
• Artificial Intelligence

Background:

• Natural hazard named entity recognition is crucial for disaster mitigation but faces challenges like entity variability.
• Existing methods struggle with the dynamic and diverse nature of natural hazard information.

Purpose of the Study:

• To develop an effective deep learning method for natural hazard named entity recognition.
• To improve the acquisition of natural hazard information for better disaster mitigation strategies.

Main Methods:

• Construction of a natural disaster annotated corpus for model training and evaluation.
• Comparison of deep learning methods utilizing word vector features, focusing on pretraining, feature extraction, and decoding.
• Proposal and implementation of the XLNet-BiLSTM-CRF model for natural hazard named entity recognition.

Main Results:

• The proposed XLNet-BiLSTM-CRF model achieved high performance with a precision of 92.80%, recall of 91.74%, and F1-score of 92.27%.
• The model demonstrated superior effectiveness compared to other evaluated methods in recognizing natural hazard named entities.
• Identified research hotspots in natural hazards literature over the past decade using the developed model.

Conclusions:

• The XLNet-BiLSTM-CRF model offers a robust and effective solution for natural hazard named entity recognition.
• Deep learning approaches can automate feature extraction, reducing reliance on manual rules in this domain.
• The method facilitates efficient information extraction, supporting natural hazard mitigation efforts.