Chinese Emergency Event Recognition Using Conv-RDBiGRU Model.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Machine Learning

Background:

• Traditional event recognition methods suffer from weak generalization and expert knowledge dependence.
• Deep neural networks face challenges with long training times and gradient dispersion.

Purpose of the Study:

• To propose a novel neural network joint model, Conv-RDBiGRU, to address limitations in traditional event recognition.
• To improve the precision, recall, and F-value of Chinese emergency event recognition.

Main Methods:

• Text corpus preprocessing including word segmentation and stop word removal.
• Utilizing word embedding to create word vector matrices.
• Extracting local semantic features via convolution and deep context features through RDBiGRU (Recurrent Deep Bidirectional Gated Recurrent Unit).
• Integrating residual structures into the recurrent neural network architecture.

Main Results:

• The Conv-RDBiGRU model demonstrated improved precision and recall for Chinese emergency event recognition.
• The F-value achieved by this method was superior to other existing approaches.
• The integration of residual structures enhanced the model's performance.

Conclusions:

• The proposed Conv-RDBiGRU model offers a more effective approach to Chinese emergency event recognition.
• Integrating residual structures into recurrent neural networks is a promising direction for improving event recognition tasks.
• This method overcomes limitations of traditional approaches and deep learning models.