Subjects in clinical trials, either patients with the target disease or healthy vohmteers, inevitably run a risk of injury or even death. To protect human subjects' rights to life and health, the Declaration of Helsi...Subjects in clinical trials, either patients with the target disease or healthy vohmteers, inevitably run a risk of injury or even death. To protect human subjects' rights to life and health, the Declaration of Helsinki has been developed as "a statement of ethical principles for medical research involving human subjects. Though widely regarded as a milestone in human research ethics, it is not a law or regulation, and is unable to effectively protect human subjects' rights. In this context, China beefs up its legal protection of clinical trial subjects.展开更多
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...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模型输入文本的长度限制,对于过长的输入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。展开更多
基金funded by the National Social Science Foundation of China (Grant No.14BFX161)
文摘Subjects in clinical trials, either patients with the target disease or healthy vohmteers, inevitably run a risk of injury or even death. To protect human subjects' rights to life and health, the Declaration of Helsinki has been developed as "a statement of ethical principles for medical research involving human subjects. Though widely regarded as a milestone in human research ethics, it is not a law or regulation, and is unable to effectively protect human subjects' rights. In this context, China beefs up its legal protection of clinical trial subjects.
文摘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模型输入文本的长度限制,对于过长的输入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。