Social networks can serve as a valuable communication channel for calls for help, offering assistance, and coordinating rescue activities in disaster. Social networks such as Twitter allow users to continuously update relevant information, which is especially useful during a crisis, where the rapidly changing conditions make it crucial to be able to access accurate information promptly. Social media helps those directly affected to inform others of conditions on the ground in real time and thus enables rescue workers to coordinate their efforts more effectively, better meeting the survivors' need. This paper presents a new sequence to sequence based framework for forecasting people's needs during disasters using social media and weather data. It consists of two Long Short-Term Memory (LSTM) models, one of which encodes input sequences of weather information and the other plays as a conditional decoder that decodes the encoded vector and forecasts the survivors' needs. Case studies utilizing data collected during Hurricane Sandy in 2012, Hurricane Harvey and Hurricane Irma in 2017 were analyzed and the results compared with those obtained using a statistical language model n-gram and an LSTM generative model. Our proposed sequence to sequence method forecast people's needs more successfully than either of the other models. This new approach shows great promise for enhancing disaster management activities such as evacuation planning and commodity flow management.