Industrial-strength schema matching

Abstract

Schema matching identifies elements of two given schemas that correspond to each other. Although there are many algorithms for schema matching, little has been written about building a system that can be used in practice. We describe our initial experience building such a system, a customizable schema matcher called Protoplasm. Schema MatchingMost systems-integration work requires creating mappings between models, such as database schemas, message formats, interface definitions, and user-interface forms. As a design activity, schema mapping is similar to database design in that it requires digging deeply into the semantics of schemas. This is usually quite time consuming, not only in level-of-effort but also in elapsedtime, because the process of teasing out semantics is slow. It can benefit from the development of improved tools, but it is unlikely such tools will provide a silver bullet that automates all of the work. For example, it is hard to imagine how the best tool could eliminate the need for a designer to read documentation, ask developers and endusers how they use the data, or review test runs to check for unexpected results. In a sense, the problem is AI-complete, that is, as hard as reproducing human intelligence.The best commercially-available model mapping tools we know of are basically graphical programming tools. That is, they allow one to specify a schema mapping as a directed graph where nodes are simple data transformations and edges are data flows. Such tools help specify: a mapping between two messages, a data warehouse loading script, or a database query. While such graphical programming is an improvement over typing code, no database design intelligence is being offered. Despite the limited expectations expressed in the previous paragraph, we should certainly be able to offer some intelligence  automated help, in addition to attractive graphics.There are two steps to automating the creation of mappings between schemas: schema matching and query discovery. Schema matching identifies elements that correspond to each other, but does not explain how they correspond. For example, it might say that FirstName and LastName in one schema are related to Name in the other, but not say that concatenating the former yields the latter. Query discovery picks up where schema matching leaves off. Given the correspondences, it obtains queries to translate instances of the source schema into instances of the target, e.g., using query analysis and data mining [5].This paper focuses on schema matching. There are many algorithms to solve it [8]. They exploit name similarity, thesauri, schema structure, instances, value distribution of instances, past mappings, constraints, cluster analysis of a schema corpus, and similarity to standard schemas. All of these algorithms have merit. So what we need is a toolset that incorporates them in an integrated package. This is the subject of this paper.The published work on schema matching is mostly about algorithms, not systems. This algorithm work is helpful, off...

Ten Challenges for Ontology Matching

“…With this respect, an important work has been conducted within the Knowledge Web project 6 . It concerned with the analysis of the Gartner hype curve 7 and placement of the various semantic web technologies along it, see Figure 3.…”

“…Beside quality of matching results, there is an issue of performance, see, e.g., [6]. Performance is of prime importance in many dynamic applications, for example, where a user can not wait too long for the system to respond.…”

Matching large schemas: Approaches and evaluation

“…Currently, only few systems, such as COMA [6], Cupid [14], Protoplasm [15], and [16], address the problem of context-dependent matching. Although based on different implementations, they resolve all shared elements and match between all their unique contexts, which we denote as the AllContext strategy.…”

Semi-automated schema integration with SASMINT

“…In addition to these projects, there are several other efforts that consider some specific features of schema matching. For example, Bernstein et al [11], Embley et al [21], Rahm et al [54]) focused on matching large schemas or extensibility of the developed schema matching system. Furthermore, the main goal of Doan et al [19], Gal [24,25], Nottelmann and Straccia [50] is to manage uncertainty in schema matching.…”