The loose and dynamic association between peers in a peer-to-peer integration has meant that, to date, peer-to-peer systems have been based on exchange of files identified with a very limited set of attributes, and no schema is used to describe the data within those files. This paper

When data sources are virtually integrated there is no common and centralized mechanism for maintaining global consistency. In consequHHj9 it is likely that inconsistencies with respect to certain global integrity constraints (ICs)will occu; In this chapter we consider the problem of defining andcompu2;) those answers that are consistent wrt the global ICs when global qubal) are posed tovirtuM data integration systems whosesou)33 are specified following the local-as-view approach.

Abstract. This paper describes the AutoMed repository and some associated tools, which provide the first implementation of the both as view (BAV) approach to data integration. Apart from being a highly expressive data integration approach, BAV in additional provides a method to support a wide range of data modelling languages, and describes transformations between those data modelling languages. This paper documents how BAV has been implemented in the AutoMed repository, and how several practical problems in data integration between heterogeneous data sources have been solved. We illustrate the implementation with examples in the relational, ER, and semi-structured data models. 1

In AutoMed, data integration is based on the use of reversible sequences of schema transformations. We discuss how views can be generated from these sequences. We also discuss techniques for optimising the views, firstly by simplifying the transformation sequences, and secondly by optimising the view definitions generated from them.

This paper describes how the AutoMed data integration system is being extended to support the integration of heterogeneous XML documents. So far, the contributions of this research have been the development of two algorithms.

In data integration, a Merge operator takes as input a pair of schemas in some conceptual modelling language, together with a set of correspondences between their constructs, and produces as an output a single integrated schema. In this paper we present a new approach to implementing the Merge operator that improves upon previous work by considering a wider range of correspondences between schema constructs and defining a generic and formal framework for the generation of schema transformations. This is used as a basis for deriving transformations over high level models. The approach is demonstrated in this paper to generate transformations for ER and relational models.

Abstract. Schema integration is the activity of providing a unified representation of multiple data sources. The core problems in schema integration are: schema matching, i.e. the identification of correspondences, or mappings, between schema objects, and schema merging, i.e. the creation of a unified schema based on the identified mappings. Existing schema matching approaches attempt to identify a single mapping between each pair of objects, for which they are 100 % certain of its correctness. However, this is impossible in general, thus a human expert always has to validate or modify it. In this paper, we propose a new schema integration approach where the uncertainty in the identified mappings that is inherent in the schema matching process is explicitly represented, and that uncertainty propagates to the schema merging process, and finally it is depicted in the resulting integrated schema. 1

Current data integration systems allow a variety of heterogeneous structured or semi-structured data sources to be combined and queried by providing an integrated view over them. The Semantic Web also requires us to be able to integrate information from a variety of heterogeneous information sources. However, these information sources will also include natural language (e.g. web pages) and ontologies. In this paper we describe an architecture which combines the data integration approach with techniques from Information Extraction in order to allow information from ontologies and natural language sources to be integrated with other, semantically related, structured or semi-structured data. Our architecture is based on the AutoMed data integration system, and in this paper we describe several extensions whichhave been made to AutoMed in order to support this work, including adding RDF to the data sources supported by AutoMed and providing a repository for the data and metadata discovered by the Information Extraction process.

Abstract. Data integration is frequently performed between heterogeneous data sources, requiring that not only a schema, but also the data modelling language in which that schema is represented must be transformed between one data source and another. This paper describes an extension to the hypergraph data model (HDM), used in the AutoMed data integration approach, that allows constraint constructs found in static data modelling languages to be represented by a small set of primitive constraint operators in the HDM. In addition, a set of five equivalence preserving transformation rules are defined that operate over this extended HDM. These transformation rules are shown to allow a bidirectional mapping to be defined between equivalent relational, ER, UML and ORM schemas. The approach we propose provides a precise framework in which to compare data modelling languages, and precisely identifies what semantics of a particular domain one data model may express that another data model may not express. The approach also forms the platform for further work in automating the process of transforming between different data modelling languages. The use of the both-as-view approach to data integration means that a bidirectional association is produced between schemas in the data modelling language. Hence a further advantage of the approach is that composition of data mappings may be performed such that mapping two schemas to one common schema will produce a bidirectional mapping between the original two data sources.

This technical report describes the XML data integration framework being built within the AutoMed heterogeneous data integration system. It presents a description of the overall framework, as well as an overview of and comparison with related work and solutions by other researchers. The contributions of this research are the development of two algorithms for XML data integration, the first for schema integration and the second for view materialization, both based on graph restructuring.