基于语法树的SAO结构识别方法研究

Parse Tree-based SAO Structure Identification

[目的/意义]SAO是一种能够表示主题信息和主题间关系的3元组结构,是文献计量学领域一个快速发展的研究方向。为了获得"满足文献计量分析需求的SAO结构",需要解决现有SAO结构识别方法遭遇的3个问题:查全和查准率低、所识别SAO结构和领域主题相关性不强以及矩阵稀疏性。[方法/过程]提出一种面向文献计量分析的基于语法树的SAO结构识别方法,首先基于共现算法和"主题词簇"方法(term clumping)识别SAO核心组件,然后利用基于语法树的抽取算法实现SAO结构的逐层抽取。[结果/结论]案例研究发现,该方法的平均查准率为0.805 8,平均查全率为0.844 6,所识别SAO结构与领域主题关系较强,且矩阵稀疏性也得到较好改善,可有效应用于相关文献计量分析。

[Purpose/significance]Subject-Action-Object （SAO） is a triple structure which can be used to both de- scribe topics in details and explore the relationship between topics. SAO analysis is a fast-growing research field. In order to obtain the SAO structures which are suitable for the bibliometric analysis, three problems need to be solved. Recall and precision have been low. The SAOs don＇ t have close relationships with domain topics. There is a problem of matrix spar- sity. [ Method/process] This paper proposed a parse tree-based SAO identification method for the bibliometric analysis. It included ： （ 1 ） a model to identify the core components of SAO structures, where co-word analysis and term clumping processes were involved; （2） a parse tree-based hierarchical SAO extraction model to implement SAO structures identifi- cation. [ Result/concluslon] The case study shows that the average precision and average recall of the proposed method is 0. 8058 and 0. 8446. The SAO extracted with our method has a great relationship with the domain topic and improves the matrix sparsity, which makes it be used as an effective tool for the bibliometric analysis.