基于密度聚类算法改进的语义主路径分析方法研究
Research on Improving the Semantic Main Path Analysis Method by Leveraging the Density Peak Clustering Algorithm
陈亮 1余池 1尚玮姣 2许海云 3吕世炅 1陈利利1
作者信息
- 1. 中国科学技术信息研究所,北京 100038
- 2. 中国林业科学研究院林业科技信息研究所,北京 100091
- 3. 山东理工大学管理学院,淄博 255000
- 折叠
摘要
语义主路径分析方法在改进传统主路径分析法中主路径内容单一、主题一致性较差等不足的同时,留下了两个缺陷,即所选主路径在语义空间的位置可能偏离主题聚簇中心、不同主路径的主题区分度并不明显.本文在语义主路径分析方法的基础上,提出一种逐步优化的主路径选择方法,即将主题聚簇密度和路径遍历权重进行叠加形成复合密度,通过调节复合密度中两个要素的比重来优化主题聚簇中心的定位,当聚簇中心的位置变化收敛后,将位于不同主题聚簇中心的路径作为结果输出.将本文方法分别用于电动汽车锂离子电池专利引文网络和材料科学领域高影响力论文引文网络,实验结果显示,本文方法所产生的多条主路径不仅在主题聚簇中的布局更加合理,而且选取不当主路径的可能性也大大降低,从而验证了本文方法的有效性.
Abstract
The semantic main path analysis(sMPA)method overcomes the shortcomings of the traditional main path anal-ysis(MPA)method,such as a single main path and low theme consistency.However,it also leaves two defects:the posi-tion of the selected main path in the semantic space may deviate from the cluster center,and the topic discrimination of dif-ferent main paths is not obvious.To address this problem,this study proposes a gradually optimized main path selection method in which topic cluster density and path traversal weight are superimposed to form a composite density,and the lo-cation of the topic cluster center is optimized by adjusting the proportion of the two elements in the composite density.When the cluster center converges,the paths located in different topic cluster centers are outputted.This method is verified by applying it to the patent citation network of lithium-ion batteries for electric vehicles and the citation network of high-impact papers in the field of materials science.The experimental results show that not only is the layout of multiple main paths generated by the new method but the possibility of selecting improper main paths is also significantly reduced.
关键词
语义主路径分析/主题一致性/主题聚类/材料科学/电动汽车锂离子电池Key words
semantic main path analysis/theme consistency/topic clustering/materials science/electrical vehicle lithium-ion battery引用本文复制引用
基金项目
中央级公益性科研院所基本科研业务费专项(ZD2023-13)
国家电网总部科技项目(1400-202357341A-1-1-ZN)
出版年
2024
NETL
NSTL
维普
万方数据