A Deep Learning-based Event Extraction Method in the Field of Electric Power Public Opinion

Abstract

Event extraction is a sub-task of information extraction in natural language processing by extracting relevant event information from unstructured text. In order to obtain the hot events related to electric power public opinion in a timely manner and assist electric power staff to make quick decisions, this article suggests a deep learning-based event extraction model for electric power public opinion, which is mainly composed of two parts, namely, an event detection model and an argumentative meta-role extraction model. The event detection model is further extracted by using the BLSTM model to obtain the specific event categories of electrical power viewpoint text, and the argumentative role extraction model is employed to extract the features of electric power opinion text by using the BLSTM-CRF model to obtain the argumentative roles included within the text. In this paper, we solve the problem of overlapping roles by using an innovative location indexing annotation method. Finally, the events contained in the power opinion text are extracted by the joint extraction of the event category and the theoretical roles. By conducting experimental tests, this research proposes a model with superior performance in terms of event extraction outcomes and accuracy rate..