Collective Ontology Alignment

    Jason B. Ellis, Oktie Hassanzadeh, Kavitha Srinivas, and Michael J. Ward

                           IBM T.J. Watson Research,
                    P.O. Box 704, Yorktown Heights, NY 10598
            {jasone,hassanzadeh,ksrinivs,MichaelJWard}@us.ibm.com



1     Introduction
Enterprises are captivated by the promise of using big data to develop new prod-
ucts and services that provide new insights into their customers and businesses.
However, there are significant challenges leveraging data across heterogeneous
data stores, including making those data accessible and usable by non-experts.
We designed a novel system called Helix which employs a combination of user-
driven and automated techniques to bootstrap the building of a unified seman-
tic model over virtualized data. Such a uniform semantic model allows users to
query across data stores transparently, without needing to navigate a maze of
data silos, data formats, and query languages.
    In this poster, we discuss a specific aspect of Helix: the method by which
it facilitates ontology alignment. Such alignments are very noisy and manually
fixing the issues is a laborious process, especially when all the work must be
done prior to putting the system into use. Instead, Helix proposes to engage
users progressively in the process of ontology alignment through the course of
their everyday use of the system. We do this by framing ontology alignment as
a guided data exploration and integration task.


2     Guided Exploration, Linking, and Sharing
Helix provides a uniform user interface for heterogeneous data exploration and
linking. This interface abstracts the underlying differences among data stores and
data representations, with the goal of allowing the user to focus on their task
rather than the technology. This interface engages users in the data alignment
process through three key features:

1. Guided navigation - search and navigate to locate results of interest, assisted
   by suggestions based on semantic and schematic links
2. Saving results - save results of interest and share those results with others
3. Guided linking - users select two saved results and Helix guides them through
   ontology alignment. The resulting linked data can be saved, shared, and used
   in future links. (see Figure 1)

    Through this process, users are creating ontology alignments by finding data
of interest, aligning it, and saving/sharing what they find useful. In this way,
2      Ellis et al.




           Fig. 1. Helix Front Page with two results selected for linking.

everyone who uses Helix is contributing to the alignment of the underlying data
sources and can leverage the work of others.
    Currently, sharing happens by one user explicitly sending saved results to
another. However, we are building a recommender system that will automatically
show relevant results to users through the course of their work, allowing them
to more readily reuse ontology alignments.
    Previous work has proposed systems that perform analysis and “pay-as-you-
go” integration in specific domains using semantic technologies [2]. However,
such systems typically leave the user out of the ontology mapping process. Helix
explicitly engages users in linking the data they are interested in.
    This work is also related to research on making it easier for non-expert users
to query standard database management systems, particularly those that take
an exploratory search approach [3]. Explorator offers a somewhat similar user
experience, allowing users to explore RDF data through a process involving
search, faceted navigation, and set operations [1]. By contrast, Helix allows users
to navigate heterogeneous data and build complex queries through a process
of progressively linking saved results. It also assists users through semantic &
schematic guidance, linkage discovery, and (ultimately) recommendations.


References
1. de Araújo, S., Schwabe, D.: Explorator: a tool for exploring RDF data through
   direct manipulation. In: Linked Data on the Web WWW Workshop (2009)
2. Lopez, V., Kotoulas, S., Sbodio, M.L., Stephenson, M., Gkoulalas-Divanis, A.,
   Mac Aonghusa, P.: QuerioCity: a linked data platform for urban information man-
   agement. In: ISWC’12: Proceedings of the 11th international conference on The
   Semantic Web. Springer-Verlag (Nov 2012)
3. White, R.W., Drucker, S.M., Marchionini, G., Hearst, M., schraefel, m.c.: Ex-
   ploratory search and HCI. In: CHI ’07 extended abstracts. pp. 2877–2880. ACM
   Press, New York, New York, USA (2007)