Comparative Analysis of RNN, LSTM, Bi-LSTM Performance for Location and Time Entity Recognition in Forest Fire Texts

This paper presents a comprehensive comparative analysis of the performance of Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Bidirectional Long Short-Term Memory (Bi-LSTM) architectures in the context of Location and Time Entity Recognition within forest fire-related texts. Leveraging a carefully curated dataset from Twitter discussions on forest fires, comprising 19,845 words and 1,103 sentences, the study employs the Beginning Inside Outside (BIO) encoding scheme for Named Entity Recognition (NER). The research explores the effectiveness of these architectures in capturing sequential dependencies and contextual nuances inherent in the informal text, such as tweets during emergencies. Implementation is carried out using TensorFlow and Keras, and the models are evaluated based on accuracy, validation accuracy, loss, validation loss, and average training time. The findings provide insights into the strengths and limitations of each model, with the Bi-LSTM emerging as a promising choice for robust and accurate Location and Time Entity Recognition in the challenging domain of forest fire texts.