A GPT-2 Based Method For Summarising Judicial Judgment Documents

The judgment of a legal case is usually very long and full of legal terms,so it is difficult for ordinary people to understand,and even professionals need to spend much time understanding.Therefore,it is necessary to summarise the judgments of legal cases.On the other hand,in re-cent years,pre-trained language models have achieved excellent results in many natural language processing tasks.In particular,the GPT-2 pre-trained language model is good at the task of text generation.And the automatic summarisation of a legal document is a kind of text generation task.Thus,this paper develops a GPT-2 based method for automatically summarising a legal judgment.Specifically,since a judgment document is lengthy,exceeding the maximum word limit for inputs into GPT-2,we first design algorithms to extract five parts of a judgment document and their coun-terparts from the document’s summary.These five parts include the dispute type of the judgment document,the claim,the defendant’s argument,the trial identification,and the judgment result part.Then we use the dataset of each part of all the judgment documents to fine-tune GTP-2.Next,for a judgment,we use each of the five fine-tuned models of GTP-2 to generate a part of the judgment summary.Finally,we put the five generated partial summaries together to obtain the overall summary of the legal document.In addition,we do lots of experiments to confirm the effectiveness of our system.