Intelligent legal judgment method based on text extraction
范阿曼 1王延川2
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 陕西国际商贸学院,陕西西安 712046
2. 西北工业大学,陕西西安 710072
折叠
摘要
深度学习在自然语言处理方面取得了巨大进展,以深度神经网络为代表的模型开始在法律智能判决上被广泛使用.基于 Transformer 的双向编码器表征法(Bidirectional Encoder Representations from Transformers,BERT)模型能够挖掘法律描述文本中双向上下文信息,利用BERT中自注意力机制完成了罪名预测、法律条款推荐、刑期预测多个司法智能审判任务.为了在长文本案情描述文本上获得更好的效果,进一步解决BERT模型输人文本的长度限制,对于过长的输人文本进行关键信息提取.在文本提取的过程中,充分利用前期训练的基于BERT智能审判模型,对于案情描述中句子的重要性进行评估,提取关键句子减少判断模型的输入长度.将精简后的案情描述文本再送入BERT模型进行司法智能审判学习.相比于直接输入原始案情描述文本的方法,基于文本提取处理后的法律描述在智能审判任务中能够取得更好的效果.
Abstract
In recent years,deep learning has achieved great success in natural language processing.The convolutional neural network has been widely used in legal intelligent decisions.Based on the Bidirectional Encoder Representations from Transformers(BERT)model,it can extract the bidirectional context information in the legal statement text,and complete multiple judicial intelligent judgment tasks such as crime prediction,legal clause recommendation and sentence prediction through the self-attention mechanism in BERT.In order to further solve the length limitation of the BERT model on the input text and obtain better results on long text legal statement samples,extract long legal statements.In the process of text extraction,the pre-trained BERT model is used to evaluate the importance of sentences in legal statements,and key sentences are extracted for redundant legal statements to reduce the input length of the judgment model and remove irrelevant information.The extracted text is sent to the BERT model for judicial trial.Compared with the original methods,the legal statement based on text extraction can achieve better results in the intelligent trial task.
关键词
法律智能审判/文本提取/多任务学习/双向编码器表征法/自然语言处理
Key words
legal intelligent trial/text extraction/multi-task learning/BERT/natural language processing
维普
万方数据