Identification and classification of transportation disaster tweets using improved bidirectional encoder representations from transformers

Social Media today has become the most relevant and affordable platform to express one's views in real-time. The #Endsars protest in Nigeria and the COVID-19 pandemic have proven how important and reliant both government agencies and individuals are on social media. This research uses tweets collected from Twitter API to identify and classify transportation disasters in Nigeria. Information such as the user, location, and time of the tweet makes identification and classification of transportation disasters available in real-time. Bidirectional Encoder Representations from Transformers (BERT) uses a transformer that includes two separate mechanisms, a decoder that produces a prediction for the task and an encoder that reads the text input. It learns contextual relations between words (sub-words) in a text. This research applied BERT with a combination of AdamW optimizers. AdamW is an improved version of stochastic gradient descent that computes an adaptive learning rate for each parameter. Our proposed model produces an accuracy of 82%. It was concluded that our approach outperformed the existing algorithm: BERT having an accuracy of 64%. © 2023

Identification and classification of transportation disaster tweets using improved bidirectional encoder representations from transformers

Social Media today has become the most relevant and affordable platform to express one's views in real-time. The #Endsars protest in Nigeria and the COVID-19 pandemic have proven how important and reliant both government agencies and individuals are on social media. This research uses tweets collected from Twitter API to identify and classify transportation disasters in Nigeria. Information such as the user, location, and time of the tweet makes identification and classification of transportation disasters available in real-time. Bidirectional Encoder Representations from Transformers (BERT) uses a transformer that includes two separate mechanisms, a decoder that produces a prediction for the task and an encoder that reads the text input. It learns contextual relations between words (sub-words) in a text. This research applied BERT with a combination of AdamW optimizers. AdamW is an improved version of stochastic gradient descent that computes an adaptive learning rate for each parameter. Our proposed model produces an accuracy of 82%. It was concluded that our approach outperformed the existing algorithm: BERT having an accuracy of 64%.