学术资源本体非等级关系抽取研究

发布时间：2018-06-15 19:08

本文选题：本体构建 + 非等级关系抽取　；参考：《图书情报工作》2016年20期

【摘要】：[目的 /意义]概念非等级关系抽取是本体构建的必要步骤,学术文献作为一种重要的学术资源类型,本文主要利用其结构特点来进行本体概念非等级关系的抽取。[方法 /过程]首先,在本体概念抽取的基础上,对文献中概念的类型进行分类,以便于后期关系动词搭配的概念类型来排除不符合条件的三元组;其次,确定学术文献中的关系类型,并采用C-value方法抽取表示关系的动词,进行本体关系的表示;再次,评价概念对的关联性,利用互信息法对概念对进行排序并去除非相关概念对,实验表明该方法非常有效;最后评价概念对与关系动词的关联,分析影响三元组关联的因素,再采用实验确定模型挖掘三元组,实验比较现有的关联规则挖掘的方法。[结果 /结论]结果表明本文提出的三元组选择模型效果明显超过现有的关联规则挖掘方法,并且在语料集扩大的情况下这种优势更为明显。
[Abstract]:Abstract of concept non-hierarchical relation is a necessary step in ontology construction. As an important type of academic resources, this paper mainly uses its structural characteristics to extract non-hierarchical relationship of ontology concept. [methods / process] first, on the basis of ontology concept extraction, the classification of concept types in literature is carried out so as to facilitate the conceptual types of post-relational verb collocations to exclude unqualified triples. To determine the types of relationships in academic literature, and use C-value method to extract verbs to express relationships, and to represent ontology relations; thirdly, to evaluate the relevance of concept pairs, and to use mutual information method to sort the concept pairs unless the related concept pairs. The experimental results show that this method is very effective. Finally, we evaluate the relationship between the concept and the relational verbs, analyze the factors that affect the association of triples, and then use the experimental deterministic model to mine the triples, and compare the existing methods of association rules mining. [results / conclusion] the results show that the proposed triple selection model is more effective than the existing association rules mining methods, and this advantage is more obvious when the corpus is enlarged.
【作者单位】：南京大学信息管理学院;
【基金】：国家社会科学基金重大招标项目“面向学科领域的网络信息资源深度聚合与服务研究”(项目编号:12&ZD221) 南京大学优秀博士研究生创新能力提升计划(项目编号:201601B033)研究成果之一
【分类号】：TP391.1

【相似文献】