Bootstrapping object coreferencing on the semantic web

Last update: 2011-12-07
URI: http://ws.nju.edu.cn/publications/hqs11
Authors: Wei Hu , Yuzhong Qu , Xingzhi Sun
Type: Article
Published: 2011
Publisher: Journal of Computer Science and Technology
For projects:
Volume: 26
Issue: 4
Pages: 662-674
Abstract:

An object on the Semantic Web is likely to be denoted with several URIs by different parties. Object coreferencing is a process to identify "equivalent" URIs of objects for achieving a better Data Web. In this paper, we propose a bootstrapping approach for object coreferencing on the Semantic Web. For an object URI, we firstly establish a kernel that consists of semantically equivalent URIs from the same-as, (inverse) functional properties and (max-)cardinalities, and then extend the kernel with respect to the textual descriptions (e.g., labels and local names) of URIs. We also propose a trustworthiness-based method to rank the coreferent URIs in the kernel as well as a similarity-based method for ranking the URIs in the extension of the kernel. We implement the proposed approach, called ObjectCoref, on a large-scale dataset that contains 76 million URIs collected by the Falcons search engine until 2008. The evaluation on precision, relative recall and response time demonstrates the feasibility of our approach. Additionally, we apply the proposed approach to investigate the popularity of the URI alias phenomenon on the current Semantic Web.

Download: file
RDF