Research on Partition Block-based Multi-source Knowledge Fusion for Knowledge Graph Construction

Authors:
Publisher website
Google Scholar
Abstract
Knowledge graph construction requires obtaining information from different data sources, of which structured data is an important source. However, due to the heterogeneity of multi-source structured data and the large size of candidate collections. The sharing and integration of knowledge is quite challenging. Therefore, how to fuse these structured data is an important task for knowledge graph construction. Based on this, the Partition Block-based Multi-source Knowledge Fusion (PB-MKF) algorithm is proposed in this paper. The PB-MKF algorithm first utilizes scalable Hierarchical Agglomerative Co-Clustering Algorithm (HACC) to achieve high cohesion of the same partition and low correlation of different partitions, then employs Dynamic Blocking Algorithm (DBA) to iteratively reduce the size of the candidate set to be matched. And the multi-strategy weighted similarity algorithm is used for entity alignment. Finally, multi-source structured knowledge fusion is realized. The PB-MKF algorithm is applied to the knowledge fusion of Chinese Obstetric Knowledge Graph (COKG), which expanded the triads from 15249 to 37831, forming an expanded COKG (ECOKG).
Funding Information
  • National key research and development program (2017YFB1002101)
  • National Natural Science Foundation of China (62006211)