电子学报 ›› 2018, Vol. 46 ›› Issue (6): 1410-1414.DOI: 10.3969/j.issn.0372-2112.2018.06.020

• 学术论文 • 上一篇    下一篇

基于加权超图随机游走的文献关键词提取算法

马慧芳, 刘芳, 夏琴, 郝占军   

  1. 西北师范大学计算机科学与工程学院, 甘肃兰州 730070
  • 收稿日期:2017-01-23 修回日期:2017-07-20 出版日期:2018-06-25
    • 作者简介:
    • 马慧芳,女,1981年7月出生,甘肃兰州人.博士,硕士生导师,现为西北师范大学计算机科学与工程学院副教授.研究领域为人工智能,数据挖掘与机器学习.E-mail:mahuifang@yeah.net;刘芳,女,1996年12月出生,甘肃平凉人.现为西北师范大学计算机科学与工程学院本科生.研究方向为文本挖掘.E-mail:1554216090@qq.com
    • 基金资助:
    • 国家自然科学基金 (No.61762078,No.61363058,No.61762079,No.61762080); 中科院智能信息处理重点实验室开放课题 (No.IIP2014-4); 广西可信软件重点实验室研究课题 (No.kx201705)

Keywords Extraction Algorithm Based on Weighted Hypergraph Random Walk

MA Hui-fang, LIU Fang, XIA Qin, HAO Zhan-jun   

  1. College of Computer Science and Engineering, Northwest Normal University, Lanzhou, Gansu 730070, China
  • Received:2017-01-23 Revised:2017-07-20 Online:2018-06-25 Published:2018-06-25
    • Supported by:
    • National Natural Science Foundation of China (No.61762078, No.61363058, No.61762079, No.61762080); Open Project of Key Laboratory of Intelligent Information Processing of Institute of Computing Technology, CAS (No.IIP2014-4); Research Project of Guangxi Key Laboratory of Trusted Software (No.kx201705)

摘要: 针对科技文献类标题短文本关键词提取时,已有自然语言处理算法难以建模文献时间与权威性且短文本词语较少建模往往存在高维稀疏问题,本文提出了一个综合实时性以及权威性的关键词提取算法为研究者进行相关推荐.该方法将文献标题视为超边,将标题中不同词项视为超点来构建超图,并对超图中的超边与超点同时加权,进而设计一种基于加权超图随机游走的关键词提取算法对文献标题的词项进行提取.该模型通过对文献来源,发表年份以及被引次数建模来对超边进行加权,根据节点之间的关联度以及每对节点在特定标题中的共现距离对超点加权.最后,通过超图上的随机游走计算出节点的重要性进而确立可推荐的关键词.实验表明,与三种基准短文本关键词提取算法相比,本文算法在精确率和召回率方面均有所提高.

关键词: 加权超图, 加权策略, 关键词推荐, 随机游走, 自然语言处理, 数据挖掘

Abstract: It is difficult for the existing natural language processing algorithms to model the time and authority of short texts such as paper titles of scientific literature.Besides,the short texts always tend to have fewer words and thus suffer from high dimension and sparsity.A keyword extraction method involving both real-time and authoritativeness is presented.A weighted hyper-graph is constructed where vertexes represent weighted terms and weighted hyper-edges measure the semantic relatedness of both binary relations and nary relations among terms.On one hand,the source of the documents,the year of publication and number of citations are considered for weighting hyper-edges,on the other hand,the degree of association between the nodes and co-occurrence distance for each pair of nodes in particular title are calculated for weighting hyper-vertexes.The random walk approach is performed on the weighted hyper-graph to obtain the recommended keywords.Experimental results demonstrated that compared with three baseline algorithms,the proposed approach is able to extract keywords with higher precision and recall.

Key words: weighted hypergraph, weighting strategy, keywords extraction, random walk, natural language processing, data mining

中图分类号: