Resolving Entity on A Large scale: DEtermining Linked Entities and Grouping similar Attributes represented in assorted TErminologies

摘要

The tremendous growth of the World Wide Web (WWW) accumulates and exposes an abundance of unresolved real-world entities that are exposed to public Web databases. Entity resolution (ER) is the vital prerequisite for leveraging and resolving Web entities that describe the same real-world objects. Data blocking is a popular method for addressing Web entities and grouping similar entity profiles without duplication. The existing ER techniques apply hierarchical blocking to ease dimensionality reduction. Canopy clustering is a pre-clustering method for increasing processing speed. However, it performs a pairwise comparison of the entities, which results in a computationally intensive process. Moreover, conventional data-blocking techniques have limited control over both the block size and overlapping blocks, despite the significance of blocking quality in many potential applications. This paper proposes a Real-Delegate (Resolving Entity on A Large scale: DEtermining Linked Entities and Grouping similar Attributes represented in assorted TErminologies) that exploits attribute-based unsupervised hierarchical blocking as well as meta-blocking without relying on pre-clustering. The proposed approach significantly improves the efficiency of the blocking function in three phases. In the initial phase, the Real-Delegate approach links the multiple sets of equivalent entity descriptions using Linked Open Data (LOD) to integrate multiple Web sources. The next phase employs attribute-based unsupervised hierarchical blocking with rough set theory (RST), which considerably reduces superfluous comparisons. Finally, the Real-Delegate approach eliminates a redundant entity by employing a graph-based meta-blocking model that represents a redundancy-positive block and removes overlapping profiles effectively. The experimental results demonstrate that the proposed approach significantly improves the effectiveness of entity resolution compared with the token blocking method in a large-scale Web dataset.