LSTM-based Trend Prediction of Public Opinion on Urban Fires

In the evolving realm of mobile technology and online self-media, understanding public opinion dynamics in urban fire incidents is crucial for effective emergency management. This study aims to analyze and forecast public opinion on urban fires, employing public opinion data from 10 urban fire incidents for a comprehensive time series analysis. This research hypothesizes that public opinion on urban fires exhibits significant temporal dynamics and utilizes the Long Short-Term Memory (LSTM) network, along with its variant, the Bi-directional LSTM (Bi-LSTM), for modeling. It evaluates their efficacy on both a relative scale and an absolute scale. The results revealed that LSTM models are found to be effective in predicting the evolution of public opinion in urban fire scenarios, especially single-layer LSTM and Bi-LSTM models with a learning rate of 0.001. The double-layer LSTM performs effectively in the incubation and diffusion stages, while singlelayer LSTM and Bi-LSTM models are more adept during the decay phase of public opinion. Given the limitations of the current datasets, future research should focus on enhancing dataset quality and diversity and refining the models' structure.