Automatic Chinese Text Summarization for Emergency Domain

Abstract

Abstract With the rapid development of the global economy, natural disasters and emergencies frequently occur. A large amount of disaster accident data and emergency case handling measures on the Internet can be used to provide technical reference and auxiliary decision-making when the various social emergency incident occurs. This study establishes an accurate Chinese text automatic short summarization model to automatically obtain summary information from accident cases. In the proposed model, Generative Pre-Training 2.0 (GPT2), which is excellent in generating tasks, is employed as the basic network structure. Adabound algorithm is used to optimize the model so that the model is not disturbed by extreme learning rates, and it converges to the global minimum at the end of training. It solves the problem that the Adam optimization algorithm causes the model to converge to the local minimum due to the extreme learning rate. Meanwhile, Jaya algorithm is utilized to optimize the hyperparameters of the Adabound for a good performance. Experimental results demonstrated that the proposed method has a significant improvement in terms of Recall-Oriented Understudy for Gisting Evaluation (ROUGE).