The Knowledge Management Lab
University of Toronto

Description

To achieve interoperability, information systems, e-commerce applications and program comprehension tools use mappings to translate data from one schema to another. Similarly, it is interoperability between data sources on the Semantic Web can be improved by defining mappings from data sources to ontologies. But both the generation and maintenance of mappings in a dynamic environment like the web are laborious, error prone and require expertise. We are attacking the schema mapping problem from two angles.

In one approach, we have constructed two tools that deal with mappings directly between schemas. One tool, Clio, generates such mappings and another tool, ToMAS, manages them. Both tools focus on mappings between any combination of XML and relational schemas. Clio allows non-expert users to provide a high-level specification of how two schemas correspond and automatically translates these specifications into semantically meaningful queries that transform data conforming to the source schema into data conforming to the target schema. The process consists of two phases. In the first phase, the high-level specifications, expressed as a set of inter-schema correspondences, are translated into a set of mappings that capture the design choices made in the two schemas. The design choices include the hierarchical organization of the data as well as schema constraints (i.e., foreign key constraints). During the second phase, these mappings are translated into queries (SQL, XQuery, or XSLT) over the source schema that generate data to populate the target schema. An important feature of the mapping algorithm is that it takes into consideration target schema constraints in order to guarantee that the generated data will not violate the integrity of the target schema.

In dynamic environments like the Web, not only may the data maintained by information sources change, but so may their schemas, semantics, and query capabilities. These changes must be reflected in the mappings. Mappings left invalid or inconsistent by such changes must be detected and updated. As large, complicated schemas become more prevalent, and as data is reused in more and more applications, manually maintaining mappings (even simple mappings like view definitions) is becoming impractical. ToMAS is a novel framework and tool for automatically adapting mappings as schemas evolve. It continuously monitors mappings and automatically detects mappings that are affected by modifications of schemas. Such mappings are then rewritten in accordance with the modified schemas. An important feature of ToMAS is that it treats mappings and schemas as first class citizens of a repository and a query language. This means that schemas and mappings can be used in queries, thus enabling their management.

In the second approach, the focus is on semantic mappings from database schemas to ontologies. Such mappings will help make the Semantic Web a reality by facilitating access to the content of legacy databases made available on the web  -- part of the "deep web". Although ontologies with rich semantics provide a way to improve interoperability between heterogeneous data sources, constructing the mappings is difficult, prone to error and may require both technical and domain expertise. As part of the MAPONTO project, we are developing a prototype interactive tool to address the problem of finding logical connections between entire database tables and domain ontologies. The tool works by first finding and expressing correspondences between table columns and ontology components, and then using these correspondences and heuristics to find logical formulas that are reasonable candidates to express the semantic connections between the ontology components and the table. We have evaluated our tool over existing relational schemas and ontologies, with good results.