关系数据库的实体间关系提取方法的研究

ENTITY RELATIONSHIP EXTRACTION IN RELATIONAL DATABASE

知识图谱需要从大量文本、图像、数据库等信息源中提取知识,而关系数据库是其中一个重要的数据源,存储了大量高质量数据。目前,有许多研究工作集中于从关系数据库到RDF的转换,主要考虑结构信息的转换,较少研究实体间语义关系的发现。提出一种基于随机森林的数据库实体间语义关系发现与转换方法,将关系数据转换为RDF,能够有效地发现数据库中实体之间的隐含语义关系。该方法构建融合数据库模式和数据内容的特征向量,设计并实现基于随机森林的实体间语义关系发现算法;基于发现的语义关系,实现多对多、一对多等实体语义关系的转换。实验结果表明,相对于传统的直接映射算法,该方法有更高的提取质量,减少了最终生成知识图谱中的冗余与错误。

Knowledge graph needs to extract information from massive text,graph,and database,and the relational database is an important data source,which stores massive high-quality data.Currently,many researches focus on transformation from relational databases to RDF,which mainly consider the structural information other than entity semantic relationship discovery.Thus,a novel entity semantic relationship discovery and transformation method of relational database was proposed based on random forest.Transforming relational data into RDF could effectively discover the hidden semantic relationship among relational databases.The feature vector of database schema and content was constructed and fused to discover the semantic relationship between entities based on random forest.The many-to-many and one-to-many entity semantic relationship transformation was realized based on discovered semantic relationships.Experimental results show that compared with traditional direct mapping method,the proposed method can achieve higher extraction quality and reduce the redundancy and error of final knowledge graph.