面向中文法律裁判文书的抽取式摘要算法

裁判文书自动摘要的目的在于让计算机能够自动选择、抽取和压缩法律文本中的重要信息，从而减轻法律从业者的工作量。目前，大多数基于预训练语言模型的摘要算法对输入文本的长度存在限制，因此无法对长文本进行有效摘要。为此，该文提出了一种新的抽取式摘要算法，利用预训练语言模型生成句子向量，并基于Transformer编码器结构融合包括句子向量、句子位置和句子长度在内的信息，完成句子摘要。实验结果显示，该算法能够有效处理长文本摘要任务。此外，在2020年中国法律智能技术评测(CAIL)摘要数据集上进行测试的结果表明，与基线模型相比，该模型在ROUGE-1、ROUGE-2和ROUGE-L指标上均有显著提升。

Extractive Summarization Algorithm for Chinese Legal Judgment Documents

The purpose of automatic judgment document summarization is to allow computers to automatically select, extract, and compress important information from legal texts so as to reduce workload of practitioners. Currently, most summarization algorithms based on pre-trained language models have limitations on the length of the input text, so they cannot effectively summarize long texts. In this thesis, an innovative extractive summarization algorithm is introduced, which uses a pre-trained language model to generate sentence vectors. Based on the Transformer encoder structure, the summarization task can be completed by fused information including sentence vectors, position and length of sentences. Experimental results showed that, the algorithm can effectively handle the task of summarizing long texts. In addition, the model was tested on the 2020 CAIL (challenge of AI in law) summarization dataset, and results showed that compared to the baseline model, the proposed model showed significant improvement in the ROUGE-1, ROUGE-2, and ROUGE-L metrics.