A Multi-Neural Network Fusion Based Method for Financial Event Subject Extraction

Abstract

Event extraction is a fundamental task in the domain of public opinion monitoring and financial risk control. Subject extraction of events with specific types is the kernel of event extraction. At present, there are some problems still existing in the mainstream event subject extraction methods, such as the inadequate use of semantic relationship between Chinese characters and the weak ability of feature learning. In order to solve these problems, this paper introduces the BERT (Bidirectional Encoder Representations from Transformers) pre-training model to enhance the semantic representation of characters, then proposes a novel event subject extraction method combing convolutional neural network (CNN) and long short-term memory (LSTM) to improve the ability of feature learning in the model. Experimental results show that the F1 score of the method proposed in this paper can reach 86.99%, which greatly improves the identification accuracy of the event subject in the financial domain.