Unrest News Amount Prediction with Context-Aware Attention LSTM

摘要

Accurately predicting social unrest events is crucial to improve public security. Currently, with the large scale news event datasets available such as GDELT, we can use the amount of unrest news to estimate the risk of instability which is particularly helpful in resource allocation and policy making. Thus in this paper we propose a context-aware attention based long short-term memory (LSTM) prediction framework named CA-LSTM to accurately predict the amount of unrest news of each country or state in the future. Specifically, we first use LSTM to learn the hidden representation from the raw time series data, and then we employ a temporal attention mechanism to learn the importance weight of each time slot. Finally, a fully connected layer is adopted to predict the future unrest news amount by combining the context information and the time series embedding vectors. We conduct extensive experiments on the GDELT data of the United States, and the results demonstrate the effectiveness of the proposed framework.