Registering Scientific Information Sources for Semantic Mediation

摘要

In a conventional information mediation scenario it is assumed that all sources, including their schemas, are known before the integrated view is defined. We have found this assumption to be unrealistic for scientific information integration - new relevant sources are discovered quite frequently, and need to be integrated incrementally with an existing federation. In this paper, we address the issue of source registration, the mechanism by which a new information source "registers" its semantics with the mediator, such that not only new views can be defined with the newly joining source, but existing views can benefit from the source without any redefinition. We approach the problem in the framework of semantic (a.k.a. knowledge-based or model-based) mediation, a version of information integration where the sources cannot be integrated solely based on their own logical schema, but need additional domain knowledge at the mediator to "glue" them together. We solve the problem by introducing a process called contextualization, whereby a source specifies a set of axioms to express its own conceptual model relative to the mediator's knowledge base. To this end, we present a context specification language CSL that allows the user to specify this mapping, and illustrate how the mediator interprets a CSL specification to update its knowledge schema and preexisting views. The examples are derived from a real-world scenario involving an ongoing collaboration with several neuroscience groups.