图模型框架下的报道性新闻自动摘要方法研究  

Research on Automatic Summary Methods for Reportable News under the Graph Model Framework

作　　者：袁琳 孙巍[1,3] 马晓敏 李周晶 项芮 Yuan Lin;Sun Wei;Ma Xiaomin;Li Zhoujing;Xiang Rui(Institute of Agricultural Information,Chinese Academy of Agricultural Sciences,Beijing 100081;Beijing Xiachu Technology Group Co.,Ltd.Beijing 100020;Key Laboratory of Agricultural Big Data,Ministry of Agriculture and Rural Affairs,Beijing 100081)

机构地区：[1]中国农业科学院农业信息研究所,北京100081 [2]北京夏初科技集团有限公司,北京100020 [3]农业农村部农业大数据重点实验室,北京100081

出　　处：《图书情报工作》2024年第17期122-135,共14页Library and Information Service

基　　金：国家重点研发计划项目“科技文献内容深度挖掘及智能分析关键技术和软件”(项目编号:2022YFF0711900);“中国农业科学院基本科研业务经费专项农业科技政策发展动向分析解读”(项目编号:Y2022ZK06)研究成果之一。

摘　　要：[目的/意义]针对现有文本自动摘要形成过程中重要技术节点——图模型框架下摘要知识表达方式中内容语义揭示深度不够的问题,提出报道性新闻自动摘要模型方案,为相关领域利用经过摘要处理后的网页报道性新闻文本数据开展实践研究提供借鉴参考。[方法/过程]利用ETM(Embedded Topic Model)融合词向量的主题模型分析工具,在图模型框架下针对目标摘要句的主题构造环节,加入主题重要度特征和语义相关性特征并重新设计报道性新闻句间统计特征,对报道性新闻文本深层次主题语义信息进行挖掘、过滤,以此初步形成报道性新闻自动摘要抽取模型;后续依据报道性新闻摘要主要功能需求提出摘要主题测度功能量化指标体系,建立测度标准与句子统计特征量化方法的对应关系,以此优化调整提出的报道性新闻自动摘要抽取模型。[结果/结论]利用图模型框架下的报道性新闻自动摘要方法具体选取农业领域科技动态报道性新闻的摘要抽取过程进行实证,建立报道性新闻自动摘要测度标准进一步得到优化后报道性新闻摘要模型方案,结果显示在外部报道性功能及内部ROUGE评价测评综合表现上优于对比方法,可以有效提高报道性新闻自动摘要抽取的准确性。[Purpose/Significance]With the graph model framework,the representation of summary knowledge is an important technical node in the automatic text summarization process.To address the issue of insufficient depth of semantic disclosure of summary content,this paper proposes a model for automatic summarization of news articles,providing a reference for practical research in related fields using summarized web reportable news text data.[Method/Process]With ETM(Embedded Topic Model),a topic model analysis tool integrating word vectors,this paper introduced topic importance and semantic relevance features into the topic construction link of the target summary sentence in the graph model framework.And it redesigned the statistical features between reportable news sentences to mine and filter the in-depth topic semantic information of the texts.Based on this,it formed the automatic summary extraction model for reportable news under the method proposed in this paper.Subsequently,according to the main functional requirement,it proposed a quantitative index system of the summary topic measurement function,and established the corresponding relationship between the measurement standard and the quantitative method to optimize and adjust the proposed model of reportable news.[Result/Conclusion]Using the graph model framework,the automatic summarization method for reportage news specifically selects the summarization process of agricultural science and technology dynamic reportage news for empirical research,establishes a measurement standard for automatic summarization of reportage news,and further obtains an optimized reportage news summarization model scheme.The results show that it performs better than the comparative method in terms of external reportage function and internal ROUGE evaluation,which can effectively improve the accuracy of automatic summarization extraction for reportage news.

关 键 词：图模型 报道性新闻自动摘要 嵌入式主题模型 ROUGE评价 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]