Instance-Based OWL Schema Matching

Abstract

Schema matching is a fundamental issue to many database applications, such as query mediation and data warehousing. It becomes a challenge when different vocabularies are used to refer to the same real-world concepts. In this context, a convenient approach, sometimes called extensional, instance-based or semantic, is to detect how the same real world objects are represented in different databases and to use the information thus obtained to match the schemas. Additionally, we argue that automatic approaches of schema matching should store provenance data about matchings. This paper describes an instance-based schema matching technique for an OWL dialect and proposes a data model for storing provenance data. The matching technique is based on similarity functions and is backed up by experimental results with real data downloaded from data sources found on the Web.

Towards a Well-Interlinked Web Through Matching and Interlinking Approaches

“…In this section, we very briefly summarise the instance-based process to create vocabulary matchings introduced in [30]. The outline of the process is as follows: S1.…”

“…Therefore, extracting values from all of their properties may lead to the wrong conclusion that the instances are not equivalent. Therefore, Leme et al [30] propose to extract values only from the matching properties of the instances.…”

“…Rather than simply using the cardinality of set intersections to define the cooccurrence matrix For instance, Leme et al [30] adopt the cosine similarity function to compare strings. The cosine similarity is defined as:…”

“…The technique extends the ontology matching process described in [30] to include complex matches between sets of datatype properties and is classified as instance-based since it depends on sets of instances.…”

“…Thus, given two sets, s and t, that contain instances of the datatype properties of the source class S and the target class T , respectively, the first phase of the technique constructs the Estimated Mutual Information matrix (EMI) [30,31] of the datatype property instances in s and t, which intuitively measures the amount of related information from the observed property instances. The scope of this phase is to identify simple datatype property matches.…”