Databases Reference
In-Depth Information
One of the contributions of OAEI was the development of a framework that iden-
tified various features of the tools, and enabled researchers to understand which
tools works best under which circumstances. We hope that a similar framework
can be developed for interactive tools, where there is an even greater variability
in capabilities and workflows supported by the tools. Some interaction and visual
paradigms only work well for small-scale ontologies, however, depending on a par-
ticular use case, these approaches may be appropriate. It would be useful to evaluate
this criteria and make such information publicly available.
The criteria for evaluation of matching tools needs to be specified. This should
include usability features, technical details about what ontologies are supported, as
well as criteria for evaluating the scalability of the approach.
Besides desktop tools, researchers are exploring web applications that make use
of crowdsourcing techniques. This paradigm introduces new directions in interac-
tion, such as social interactions between users, interactions to upload and share
ontologies, and services for consuming the matchings. This is a growing research
direction and it will take time to determine how to motivate users to contribute to
such projects. Also, evaluation will be important to help determine the quality of
matchings that are contributed in this way, compared to more closed settings.
Such an approach is very attractive given the success of many existing crowd-
sourcing applications. This technique is one possible approach for helping deal with
the scalability issue of generating a matching. It is a difficult and time-consuming
process for a single individual to create the entire matching between two large
ontologies. Crowdsourcing potentially alleviates some of this burden by allowing
any Web user to contribute.
Researchers who work on the tools for interactive ontology matching, must focus
more attention on the issues of scalability of the tools. As the sizes of the ontologies
grows (e.g., some biomedical ontologies have tens of thousands of classes), so do the
computational demands on the tools: they must be able to work with ontologies that
may not load into memory or may take huge computational resources to process.
Scalability of visualization techniques is another issue that must be addressed by
the tools. As the ontologies become larger, some of the visualization paradigms
that worked very well for small ontologies, with all the classes fitting on a single
computer screen, simply may not work for ontologies where only a small fraction
of the classes will fit on the screen. Both incremental matching [
Bernstein et al.
2006
] and ontology modularization [
Stuckenschmidt et al. 2009
] are approaches that
potentially address this problem. They have the potential to help reduce cognitive
overload during the matching process by restricting the focus of the user to particular
areas of the ontology.
Finally, we still must explore new questions in interactive ontology matching,
such as how to match the expertise of the user with particular areas of the ontol-
ogy, where the best location to begin a matching process is, and how to best locate
candidate-heavy areas of two ontologies.
