Yuanchun Wang

Yuanchun Wang

Auto-generated profile, based on publicly available information

Yuanchun Wang

The 54th Research Institute of CETC, China

Research Focus & Metrics

Representative Publications

View All Publications
27 October 2023
Verified
Conference Paper
Open access
BQBaohua Qiang
Baohua Qiang
FTFeng Tang
Feng Tang
XZXiangyu Zhou
Xiangyu Zhou
YWYuanchun Wang
Yuanchun Wang
XYXianyi Yang
Xianyi Yang
KLKai Li
Kai Li
To address the problems of inaccurate event element extraction, inability to directly handle long texts and errors in the fine-tuning stage of the task due to insufficient semantic information and incomplete semantic feature samples in the current Chinese domain event extraction model, this paper proposes the Chinese Event Extraction Algorithm of Multi-Information Semantic Enhancements (MiSE). In the embedding stage, the algorithm is designed a coding structure that fusing multiple information: Using the Roformer model based on rotational encoding for word embedding and text embedding, and augment the Roformer word embedding matrix by mapping the word embedding matrixs of event trigger, argument information and their location information to it , forming an vector matrix that incorporates multiple information; the matrix is then placed into the BiLSTM for forward and backward chain calculation to enhance the model's ability to extract semantic information from Chinese text. In the decoding stage, using the decoding structure to obtain the set of tags with the highest probability. Comparing this algorithm with several Chinese event extraction algorithms on the Baidu DuEE dataset and its extraction results outperformed other event extraction algorithms in all evaluation metrics.