We replicate and demonstrate that the performance of the WikiMatch automated ontology alignment system may be driven not by the particular information from Wikipedia directly used by the system, but rather by string similarity and Wikipedia's manually curated synonym sets, as encoded in the site's query resolution and page redirection system. In order to gain a detailed understanding of how Wikipedia contributes to WikiMatch, we replicate results reported for WikiMatch and analyze the results to evaluate our hypothesis.
Introduction
The idea behind WikiMatch is to use Wikipedia's general search functionality
(through the MediaWiki API1) to retrieve a list of related article titles for each
of the entities in the two ontologies to be aligned. After retrieving the list of
titles, the similarity of each pair of entities is computed by the Jaccard index2
on these titles. If the similarity exceeds a threshold, WikiMatch considers the
entities equivalent. We began our WikiMatch replication e ort by downloading
the source code from the link speci ed in [
Precision Recall F-measure TP FP FN 0.74 0.49 0.58 150 52 155 0.74 0.49 0.58 150 52 155 0.70 0.50 0.58 152 64 153 0.99 0.62 0.77 937 11 579 0.99 0.62 0.77 947 11 569 0.96 0.64 0.77 966 43 550
Acknowledgments This work was supported by the US Geological Survey agreement G16AC00120 \Demonstration of Semantic Web Technologies as Applied to Surface Water Feature Classi cation."