基于词、句、实体协同的关键实体抽取算法  

Key Entity Extraction Based on Word-Sentence-Entity Co-Ranking

作　　者：刘媛媛 史佳欣 李响 李涓子[2] LIU Yuanyuan;SHI Jiaxin;LI Xiang;LI Juanzi(State Key Laboratory of Mathematical Engineering and Advanced Computing,Zhengzhou 450001,China;Knowledge Engineering Group,Department of Computer Science and Technology,Tsinghua University,Beijing 100084,China;State Key Laboratory of Intelligent Technology and Systems,Tsinghua University,Beijing 100084,China)

机构地区：[1]数学工程与先进计算国家重点实验室,河南郑州450001 [2]清华大学计算机科学与技术系知识工程研究室,北京100084 [3]清华大学智能技术与系统国家重点实验室,北京100084

出　　处：《信息工程大学学报》2020年第3期304-309,共6页Journal of Information Engineering University

基　　金：国家自然科学基金资助项目(U173620030)。

摘　　要：关键实体是指单篇文档中较为重要、与文档主题关联度较大、可以概括文章主题的实体。关键实体的抽取可以帮助搜索引擎和问答系统去除无关实体,更加精准地匹配用户需求。目前已有的方法一般只考虑来自实体层次的同质影响关系,忽略了词、句等异质影响关系。因此提出Multi-Rank,一种词、句、实体协同的计算框架,利用词、实体、句子之间的互增强关系,迭代计算实体的重要度,并从理论上证明了迭代过程的收敛性。实验表明,相较于经典TextRank算法,该算法在准确率、召回率指标均上升了2%,提高了关键实体的抽取质量。Key entities refer to the entities that are more important in the document,they have great relevance to the subject and can summarize the subject of a document.Key entity extraction can help search engines and question answering systems remove irrelevant entities and match users’needs more accurately.The existing methods generally only consider the homogenous influence from the entity layer,and ignore the heterogeneous influence from the word and sentence layer.We propose Multi-Rank,a word-sentence-entity co-ranking algorithm.It integrates the mutual reinforcement between words,sentences and entities,and iteratively computes the salience of entities,we also prove the convergence.The results of the experiment show that,compared with the classic TextRank,our method improves the precision and recall by 2%,which demonstrates the effectiveness of the proposed method.

关 键 词：实体抽取 互增强关系 图排序 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]