Unveiling the Power of Deep Learning: A Comparative Study of LSTM, BERT, and GRU for Disaster Tweet Classification

초록

Disasters have serious effects on people's lives and buildings. Therefore, social media platforms, such as Twitter, have become more critical. They are crucial tools for responding to and managing disasters effectively. This study examined the effectiveness of various deep learning models, such as bidirectional encoder representations from transformers (BERT), gated recurrent units (GRU), and long short-term memory (LSTM) for classifying disaster-related tweets. Twitter data related to different disasters were collected using hashtags. The data were then cleaned, preprocessed, and manually annotated by a team. The annotated data were divided into training, validation, and testing sets. The data were used to train three models based on BERT, GRU, and LSTM for the categorical classification of disaster tweets. Finally, the three models were evaluated and compared using the test data. BERT achieved an accuracy of 96.2%, making it the most effective model. In contrast, the LSTM and GRU models achieved an accuracy of 93.2% and 88.4%, respectively. These findings underscore the potential effectiveness of deep learning models in classifying disaster-related tweets, offering insights that could enhance disaster management strategies, refine social media monitoring processes, bolster public safety, and provide directions for future research. © 2023 Institute of Electronics and Information Engineers. All rights reserved.

키워드