An efficient method for disaster tweets classification using gradient-based optimized convolutional neural networks...

Area of Science:

• Natural Language Processing
• Machine Learning
• Deep Learning

Background:

• Microblogging platforms like Twitter are crucial for disaster communication.
• Classifying real vs. fake disaster-related tweets is vital to prevent chaos.
• Traditional machine learning models show limited accuracy in tweet classification.

Purpose of the Study:

• To propose an efficient deep learning model for classifying disaster-related tweets.
• To enhance tweet classification accuracy by integrating BERT embeddings with CNN.
• To optimize the CNN model using gradient-based optimizers, specifically RMSprop.

Main Methods:

• A Convolutional Neural Network (CNN) was selected as the primary classification model.
• BERT embeddings were incorporated to capture contextual semantics.
• The model was optimized using various gradient-based optimizers, with RMSprop yielding the best results.

Main Results:

• The proposed BERT Embedding based CNN model with RMSProp Optimizer achieved an F1 score of 0.80.
• The model demonstrated a high accuracy of 0.83 in classifying disaster tweets.
• Comparative analysis validated the superiority of the proposed deep learning approach over traditional methods.

Conclusions:

• The BERT Embedding based CNN model with RMSProp Optimizer is an effective method for classifying disaster-related tweets.
• Leveraging BERT embeddings significantly improves the model's ability to understand nuanced language in disaster scenarios.
• This research contributes a robust solution for real-time disaster information management on social media.