<![CDATA[Background: Social media has become one of the primary sources of disaster-related information; however, the rapid dissemination of heterogeneous, duplicated, and misleading content presents significant challenges for retrieving reliable information that can support disaster management and educational awareness. Conventional machine learning approaches often struggle to simultaneously capture semantic representations and contextual dependencies within large-scale disaster communications.Aims: This study aims to develop and evaluate an integrated Deep CNN–LSTM-based disaster management framework for retrieving reliable disaster-related information from heterogeneous online sources while supporting educational awareness through safe information dissemination.Methods: A quantitative experimental design was employed using disaster-related textual datasets collected from multiple online platforms, including earthquake, flood, wildfire, and COVID-19 events. The proposed framework integrated Deep Convolutional Neural Networks (Deep CNN) for semantic feature extraction with Long Short-Term Memory (LSTM) networks for sequential contextual learning. Model performance was compared with six benchmark machine learning and deep learning approaches, namely Support Vector Machine (SVM), Neural Network (NN), Feed-forward Neural Network (FFNN), Recurrent Neural Network (RNN), LSTM, and Hybrid CNN–LSTM. The framework was evaluated using retrieval performance together with Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²).Result: The proposed framework consistently retrieved a greater amount of relevant disaster-related information across multiple disaster scenarios than the benchmark models. Quantitative evaluation further demonstrated robust predictive capability, achieving an RMSE of 15,286.2154, an MAE of 13,789.75, and the highest coefficient of determination (R² = 0.9793), indicating strong model fitting and reliable predictive performance.Conclusion: The integrated Deep CNN–LSTM framework effectively combines semantic feature extraction and contextual sequence learning to improve disaster information retrieval from heterogeneous online sources. The proposed framework contributes to intelligent disaster management by providing reliable disaster-related information that supports educational awareness and evidence-based decision-making during disaster preparedness and emergency response.]]>
