PLE-MobileBERT: enhancing call content classification during emergencies using the ECP dataset

Researchers have been concentrating on categorizing text into several categories during the past few years based on tweets, SMS messages, etc. This study focuses on categorizing the justified calls based on call content from India’s emergency response support systems. The method becomes substantially more challenging because of the close relationship between certain emergence classes and the scarcity of resources on call transcripts. This work uses five labels for classification: Unintentional_Injury_Event, Event_Related, Medical_Emergency, Aid_Related, and Crime_Related. This research performs a variety of tests to classify emergency-related calls. We employed established machine learning techniques and transformer-based NLP models, like AlBERT, RoBERTa, and MobileBERT. The proposed PLE-MobileBERT approach outperforms other models in terms of speed and effectiveness, adding the model’s ability to understand phrase-level information. The proposed PLE-MobileBERT provides an accuracy: 95.31% percent and a macroaverage F1 score: 94.8%, which is superior to the others.