Chinese Event Detection Combining BERT Model with Recurrent Neural Networks

Abstract

With the rapid development of computer technology and the scale of the Internet, it is particularly important to extract useful information from the increasing amount of information on the Internet. Among them, event extraction is one of the research hotspots in the field of natural language processing and an important research direction in the field of information extraction. Event detection is the first step of the event extraction task, which plays a decisive role in the subsequent event extraction work. The article adopts BERT for word vector training, joint lexical vector, named entity vector, and semantic dependency vector as the input of Bi-LSTM, and then input them into the CRF sequence tagging layer after acquiring the features of sentences to achieve the recognition of event trigger words and the classification of event categories. The CEC corpus is selected as the training and test set, and experiments show that the method is effective in event detection, with F-values up to more than 70%.