PORSCHE: Performance ORiented SCHEma Matching
Abstract
Semantic matching of schemas in heterogeneous data sharing systems is time consuming and error prone. Existing mapping tools employ semi-automatic techniques for mapping two schemas at a time. In a large-scale scenario, where data sharing involves a large number of data sources, such techniques are not suitable. In this paper we present a method, which creates a mediated schema tree from a large set of input schema trees and defines mappings from the contributing schemas to the mediated schema. It is a two-phase approach. First, we use a set of linguistic matchers, which extract the semantics of all distinct node labels, present in input schemas, and form clusters of semantically similar labels. Second, we use a tree-mining data structure, combined with the similar label clusters, to calculate the context of each node, which is used in mapping. Since the input schemas are trees, our tree mining algorithm uses node ranks calculated by pre-order traversal. Tree mining combined with semantic label clustering minimizes the target search space and improves performance, thus making it suitable for large scale data sharing. We report on experiments with up to 80 schemas containing 83,770 nodes. PORSCHE took 587 seconds to match and merge them to create a mediated schema and to return mappings from input schemas to the mediated schema. We compare the quality of matching of PORSCHE with COMA++ on standard XML schemas, and find them to be very similar to the mappings produced by COMA++.