基于中文学术文献的领域本体概念层次关系抽取研究

Learning Concept Hierarchies from Chinese Academic Literature for Domain Ontology Construction

基于学术文献构建领域本体对促进领域学科发展具有重要的意义。本文提出了一种以中文学术文献为数据源,半自动化抽取领域本体层次关系的框架方法。首先,构建了一个通用的领域本体层次关系的细粒度研究框架。其次,设计了一种新的概念表示方法,融合了深度学习方法得到的概念语义特征和上下文的时间序列词频。进一步结合了AP聚类、Prim算法和Web搜索引擎的查询数据,提出了基于规则推理的本体概念层次关系抽取算法(RROCHE),实现了半自动化概念层次关系抽取。最后,基于中文分词领域的中文学术文献数据,通过数值实验方法讨论了方法的可行性和有效性。本文提出的框架方法也非常容易推广并应用到各领域本体层次关系任务中。

Constructing domain ontology from academic literature has great significance in promoting discipline development. Taking Chinese academic literature as a data source, this study proposed a semi-automatic method for extracting concept hierarchy. First, a fine-grained universal research framework for constructing hierarchical relations of the domain ontology was proposed. Then, a novel concept representation fusion method was developed, considering concepts semantic features based on deep learning and concept frequency in time series. Combined with an affinity propagation(AP) clustering algorithm, Prims algorithm, and data from a Web search engine, the ontology concept hierarchy extraction algorithm was proposed via rule-based reasoning(RROCHE). Concept hierarchy relations are learned semi-automatically. The algorithm was then applied to the academic literature on Chinese word segmentation. Numerical experiments examined the feasibility and effectiveness of the proposed methods. The proposed method can also be applied effectively and widely to other domains.