我国是受台风灾害威胁最为严重的国家之一，及时、动态地掌握台风灾害事件信息，对台风灾害开展快速地灾害损失估计和应急物资需求预测，能够为科学应急提供支撑和依据。社交媒体数据具有更新频率快、传播途径多、用户参与程度广泛的特点，在灾害应急中显示出了很好的应用潜力。本文面向台风灾害的灾害损失估计和应急物资需求预测，基于多种深度学习方法，开展了社交媒体谣言检测和数据可信度分析，融合社交媒体文本数据、气象数据、地理数据和社会经济数据等，构建了台风灾害经济损失评估模型和应急物资需求预测模型。主要研究内容如下：台风灾害下社交媒体数据可信度分析。针对社交媒体推文的用户信息、文本特征和传播结构特征，建立了结合前后向长短时记忆网络（Bi-LSTM）、图卷积神经网络（GCN）和注意力机制的社交媒体多特征融合谣言检测模型，揭示了台风灾害下的社交媒体谣言传播规律，并对台风灾害下的社交媒体数据可信度进行了分析。融合社交媒体文本数据的台风灾害损失评估。构建了基于文本卷积神经网络（TextCNN）的台风灾害下社交媒体文本分类方法，揭示了社交媒体信息与台风灾害损失之间的关联关系。结合致灾因子、孕灾环境、承灾体和社交媒体数据，建立了融合社交媒体文本分类数据的台风灾害损失评估模型，实现了台风灾害下城市尺度的直接经济损失快速评估，并通过与实际台风灾害损失数据对比，对模型进行了验证。基于社交媒体文本数据挖掘的台风灾害应急物资需求预测。构建了结合条件随机场的前后向长短时记忆网络（Bi-LSTM-CRF）和特征词匹配融合的台风灾害下社交媒体数据灾情信息提取方法。提出了社交媒体数据灾情信息和空间信息扩散相结合的受灾区域应急物资需求预测模型，并以典型台风案例开展了应急物资需求快速预测，与实际应急物资调配数据对比，对模型进行了验证。本文基于多种深度学习方法，构建了社交媒体谣言检测模型，并融合社交媒体数据构建了台风灾害损失评估模型和应急物资需求预测模型，为台风灾害应急提供决策参考。
China is one of the countries most seriously threatened by typhoon disasters. Timely and dynamic learning about typhoon disaster event information, rapid disaster loss estimation, and emergency material demand forecasting for typhoon disasters can provide support and a basis for scientific emergency response. Social media data, characterized by its rapid update frequency, multiple dissemination channels,?and extensive user participation, holds great potential for application in disaster emergency response.This paper focuses on estimating disaster losses and predicting?emergency material demand during typhoon disasters. It utilizes various deep learning methods to detect misinformation in social media posts and analyze data credibility. The study integrates social media text data, meteorological data, geographic data, and socio-economic data to develop models for assessing economic losses from typhoon disasters?and predicting emergency material demand. The main research contents are as follows:Trustworthiness analysis of social media data during typhoon disasters. A multi-feature fusion rumor detection model for social media is developed, which combines Bidirectional Long Short-Term Memory Networks (Bi-LSTM), Graph Convolutional Neural Network (GCN), and an attention mechanism. This model focuses on?user information, textual features, and dissemination structure characteristics of social media tweets. It aims to uncover the rumor dissemination patterns in social media during typhoon disasters and assess the credibility of social media data in such situations.Typhoon disaster damage assessment by fusing social media text data. A method for classifying?social media text based on Text Convolutional Neural Network (TextCNN) during typhoon disasters is developed, revealing?the correlation between social media information and losses caused by typhoon disasters. A typhoon disaster loss assessment?model was developed by integrating social media text classification?data. This model combines?disaster-causing factors, disaster-conceiving environment, disaster-bearing?bodies,?and social media?data. It enables?the rapid assessment of direct economic losses at the urban scale?during?typhoon?disasters. The model‘s validity?was confirmed?by comparing it?with actual typhoon disaster loss data.Typhoon?disaster emergency material demand prediction based on social media text data mining. A method for extracting disaster information from social media data?during?typhoon?disasters is developed by?combining the Bidirectional?Long Short-Term Memory?Network with?Conditional Random Field?(Bi-LSTM-CRF) and feature word matching?fusion.?An emergency material demand prediction model?that integrates?social media?data,?disaster?information,?and spatial information diffusion for the affected area is?proposed. A?rapid prediction of emergency material demand is?conducted using?a typical typhoon case, and the model is validated by?comparing it?with the actual emergency material deployment data.This paper presents a model for detecting false information on social media using multiple deep learning methods. It also integrates social media data structures to develop a model for assessing typhoon disaster damage and predicting emergency material demand. These models offer valuable insights for?decision-making?in typhoon disaster emergency management.
