摘要
[目的/意义]引入时间衰减因子与聚类系数对共词网络链路预测局部相似性指标进行优化改进,以期进一步提升共词网络链路预测的精确度。[方法/过程]从3个角度来提升局部相似性指标预测精确度:一是引入时间衰减因子计算连边权重,对不同时间段的连边进行区别处理;二是结合聚类系数改进已有相关相似性指标,即利用节点邻域拓扑信息进一步区分不同节点对相似性的贡献;三是同时融合时间衰减因子和聚类系数进行指标优化改进。再以图书情报学领域为例构建共词网络,利用多种分类算法(如朴素贝叶斯、决策树、随机森林、SVM)预测和评估所改进指标的预测精确度。[结果/结论]引入时间衰减因子,指标WCN、WAA、WPA、WRA的预测精确度均得到了有效提升;结合聚类系数,指标CN、AA、RA、WCN、WPA、WRA和SA的预测精确度均得到了不同程度的提升;同时融合时间衰减因子和聚类系数,WCN、WAA、WRA的预测精确度得到了进一步的提升。由此可知,引入时间衰减因子或聚类系数均能在一定程度上提升共词网络链路预测相关指标的准确度。
[Purpose/significance]This paper introduces time decay factor and clustering coefficient to optimize and improve the link prediction indexes in co-word network,so as to further improve the accuracy of link prediction in co-word network.[Method/process]This study tries to improve the prediction accuracy of local similarity indexes from three perspectives.First,a time decay factor is introduced to calculate the edge weight,and the edges generated in different time periods are processed differently.Secondly,the clustering coefficient is combined to improve the existing related indexes,and the contribution of different nodes to similarity can be further distinguished based on the topology information of nodes’neighborhood.Thirdly,both the time decay factor and the clustering coefficient are used to optimize and improve the link prediction indexes in co-word network.On this basis,the field of library and information science is taken as an example to construct a co-word network.And then a variety of classification algorithms(such as Naive Bayes,Decision Tree,Random Forest,SVM)is utilized to predict and evaluate the prediction accuracy of the improved indexes.[Result/conclusion]The accuracy of WCN,WAA,WPA and WRA was improved effectively after the time decay factor was introduced.Combined with the clustering coefficient,the accuracy of indexes CN,AA,RA,WCN,WPA,WRA and SA all improved to different degrees.When the time decay factor and the clustering coefficient are used together.The accuracy of WCN,WAA and WRA was further improved.in summary,the introduction of time decay factor or clustering coefficient can improve the accuracy of co-word network link prediction.
出处
《情报理论与实践》
CSSCI
北大核心
2022年第7期165-173,共9页
Information Studies:Theory & Application
基金
国家社会科学基金一般项目“时间感知的个性化学术文献引文推荐研究”的成果之一,项目编号:21BTQ072。
关键词
共词网络
链路预测
时间衰减因子
聚类系数
局部相似性指标
co-word network
link prediction
time decay factor
clustering coefficient
local similarity indexes
