Leveraging Social Media Posts for Natural Disaster Response Using CNN-BiLSTM Model

In the contemporary digital era, social media, particularly Twitter, has become a vital channel for sharing realtime information on natural disasters like floods, forest fires, and earthquakes. This rapid dissemination enables swift responses to disasters. However, classification models relying solely on social media data face challenges, notably the lack of richness and diversity in word embedding vectors, limiting precise analysis. To address this, the research integrates data from Wikipedia to improve word embeddings. The classification model employs Word Embedding, Convolutional Neural Network (CNN) 1D, and Bidirectional Long Short-Term Memory (BiLSTM) techniques. By combining embedding methods such as Word2Vec, GloVe, and FastText, derived from both social media and Wikipedia Indonesia, the model enhances word representation and improves classification accuracy. Results demonstrate accuracies of 85.61% for floods, $92.56 \%$ for forest fires, and $84.11 \%$ for earthquakes. This research contributes to advancing natural disaster communication classification by utilizing more diverse data and proposing methodologies for unstructured social media content analysis.