Research On Event Extraction For Judgment Documents

The information contained in the judgment documents is rich and valuable,and the structure is relatively simple compared with other text information on the Internet.It is one of the current research hotspots;event extraction is an important research direction in the field of information extraction,and it is also the most challenging research direction that has always attracted the attention of many researchers and institutions.This paper uses the judgment documents as the corpus to conduct the event extraction research.Through the event extraction technology,the judgment documents can be presented highly concisely.It only takes a few lines to understand the content of the entire judgment document,which greatly improves the efficiency of judicial personnel.It has laid a solid foundation for building a knowledge map in the field of judgment documents and building a judgment result prediction system.The main work of this paper is as follows:(1)Established an event extraction corpus based on traffic accident judgment documents.By analyzing the wording characteristics of the judgment documents,then performing word segmentation,removing stop words and other operations,and then carrying out corresponding annotations,and finally forming a small judgment document event extraction corpus.(2)Judgment document event extraction model based on hybrid neural network.First use the pre-trained BERT model to obtain the word embedding vector,then use the BiGRU-CRF model to complete the feature extraction of the character vector,combine the extraction result with the distance vector,and input it into the CRF role assignment model to effectively solve the event argument role For problems with many types of tags and uneven distribution,complete the extraction of judgment documents and events.(3)Judgment document event extraction model based on multi-feature fusion.The BERT model is used to capture the context information of word vectors,and a word vector representation method including semantic and grammatical information of words is proposed for trigger word classification tasks.The word vectors in the sentence are input in order,and the BiGRU model is used to obtain the output vector containing the sentence-level context information;all the output vectors of BiGRU are input into the Softmax classifier to realize the recognition of trigger words,and then use a CRF model implements the assignment of argument roles.The experimental results show that the hybrid neural network model effectively improves the performance of the event extraction of the referee document;the performance of the multi-feature fusion method is better than the baseline method and the hybrid neural network model,and has achieved satisfactory results.