PBA: Partition and blocking based alignment for large knowledge bases

Abstract

The vigorous development of semantic web has enabled the creation of a growing number of large-scale knowledge bases across various domains. As different knowledge-bases contain overlapping and complementary information, automatically integrating these knowledge bases by aligning their classes and instances can improve the quality and coverage of the knowledge bases. Existing knowledge-base alignment algorithms have some limitations: (1) not scalable, (2) poor quality, (3) not fully automatic. To address these limitations, we develop a scalable partition-and-blocking based alignment framework, named Pba, which can automatically align knowledge bases with tens of millions of instances efficiently. Pba contains three steps. (1) Partition: we propose a new hierarchical agglomerative co-clustering algorithm to partition the class hierarchy of the knowledge base into multiple class partitions. (2) Blocking: we judiciously divide the instances in the same class partition into small blocks to further improve the performance. (3) Alignment: we compute the similarity of the instances in each block using a vector space model and align the instances with large similarities. Experimental results on real and synthetic datasets show that our algorithm significantly outperforms state-of-art approaches in efficiency, even by an order of magnitude, while keeping high alignment quality.