基于机器阅读理解的新闻时间线挖掘与展示

News Timeline Mining and Presentation Based on Machine Reading Comprehension

[目的/意义]为了解决互联网时代信息过载的问题,尝试从海量的新闻报道中自动抽取新闻事件,并按照事件发生的时间进行排序,向用户展示结果。[方法/过程]利用网络爬虫自动获取新闻文本,使用中心事件句判定和RoBERTa+MRC的方法进行实体抽取,从非结构化文本中抽取语义信息并生成结构化的新闻摘要。在此基础上,对摘要中的时间信息进行标准化处理,并按照时间顺序生成新闻时间线。[结果/结论]实验结果表明,中心事件句判定和RoBERTa+MRC相较于传统方法有更好的事件抽取效果,文章设计的新闻时间线展示系统能够帮助用户了解新闻事件的发展脉络,解释新闻主题的事件走向与演化规律。

[Purpose/significance]In order to solve the problem of information overload,this paper attempts to automatically extract news events from a large number of news reports and show them to users according to chronological order.[Method/process]We use web crawler to automatically obtain the raw news,then entity extraction is performed using the methods of prime sentence determination and Roberta-MRC.Semantic information is extracted from unstructured content and generates structured news summary.On this basis,time in the summary is normalized,and the news timeline is generated in chronological order.[Result/conclusion]The experimental results show that prime sentence determination and Roberta-MRC have better entity extraction effect than the traditional method.The timeline system can also help users quickly understand a news topic and improve the efficiency of information acquisition.