基于深度学习的多维度证据要素关联关系抽取研究

Research on Extraction of Multi-dimensional Evidence Elements Association Relations Based on Deep Learning

随着社会法治化发展,法院受案数量呈爆发式增长,各证据要素之间关联较弱,影响法律审判效率。基于上述问题,提出一种基于深度学习的多维度证据要素关联关系抽取模型,设计基于BERT_BiGRU_CNN模型的证据要素关联关系抽取算法。本文以裁判文书为数据集,使用BERT训练所需字词向量,开展BERT、BERT_CNN、BERT_BiGRU以及BERT_BiGRU_CNN的对比研究,研究关系抽取技术在证据要素间的识别效果、研究表明,采用BERT_BiGRU_CNN算法在多要素证据关联关系之间的识别效果最好,且相比于其他模型,BERT_BiGRU_CNN的F1得分达到84.3%。

With the development of the rule of law in society,the number of court cases has exploded,and the correlation between the various elements of evidence is weak,which affects the efficiency of legal trials.Based on the above problems,a multi-dimensional evidence element association relationship extraction model based on deep learning is proposed,and an evidence element association relationship extraction algorithm based on the BERT_BiGRU_CNN model is designed.This paper uses judge documents as a data set and uses the word vectors required for BERT training to carry out a comparative study of BERT,BERT_CNN,BERT_BiGRU and BERT_BiGRU_CNN,to study the recognition effect of relation extraction technology in legal materials.Research shows that the BERT_BiGRU_CNN algorithm has the best recognition effect between multi-element evidence associations,and compared with other models,the F1 score of BERT_BiGRU_CNN reaches 84.3%.