=Paper=
{{Paper
|id=None
|storemode=property
|title=SemantEco Annotator
|pdfUrl=https://ceur-ws.org/Vol-1035/iswc2013_demo_41.pdf
|volume=Vol-1035
|dblpUrl=https://dblp.org/rec/conf/semweb/SeyedCALLPM13
}}
==SemantEco Annotator==
https://ceur-ws.org/Vol-1035/iswc2013_demo_41.pdf
SemantEco Annotator
Patrice Seyed1,2, Katherine Chastain2, Brendan Ashby2, Yue Liu2,
Timothy Lebo2, Evan Patton2, and Deborah McGuinness2
1DataONE, University of New Mexico, Albuquerque, NM 87131
2Tetherless World Constellation, Rensselaer Polytechnic Institute, Troy, NY 12180, USA
{seyeda2, chastk, ashbyb, liuy18, lebot, pattoe, dlm}@rpi.edu
Abstract. Generating useful RDF linked data is not a straightforward process
for scientists using today’s tools. In this paper we introduce the SemantEco
Annotator, a semantic web application that leverages community-based
vocabularies and ontologies during the translation process itself to ease the
process of drawing out implicit relationships in tabular data so that they may be
immediately available for use within the LOD cloud. Our goal for the
SemantEco Annotator is to make advanced RDF translation techniques
available to the layperson.
1 Introduction
Scientists generating datasets of tabular data make choices about how they record that
data. These decisions are informed by many factors, including how scientists
understand and analyze the data. Often, information that can be gleaned from a table
alone is limited by the initial input strategy, including what is appropriate for the table
headers (i.e., attributes) and cell values, and the conventions for structure and
terminology. There may be implicit information apparent only to the creator, leading
to challenges for uniformly interpreting data generated by different researchers.
Linked data formats based on web standards such as RDF provide an avenue for
addressing this problem, allowing the data to be more explicitly represented in
statements uniquely typing entities and relationships described in data using common
vocabularies. Unfortunately, the process for translating source data into linked data
has its own barriers, as software tools allowing scientists that are inexperienced with
linked data to perform translations are lacking.
In our demonstration we introduce SemantEco Annotator, a semantic web
application for translating tabular data into RDF for immediate use in the LOD cloud.
The annotator serves as a frontend to the csv2rdf4lod [1] conversion tool that uses an
RDF metadata vocabulary to specify a variety of powerful techniques for translating
data from CSV to RDF. We see the SemantEco Annotator as an application that
makes translation more accessible to the layperson, and which can act as a module in
a larger semantic workbench. We created our initial implementation for our
SemantEco environment, which uses linked data to present water quality data in a
visual and interactive manner [2]. In the next section we describe other RDF
translation tools, following which we highlight the primary translation techniques the
annotator currently enables, and finally we provide some insight into how the
application was developed, and our future work.
2 Related Work
There are existing tools both for converting tabular data to RDF and for semantic
annotation. RDF Refine1 is an extension to OpenRefine where conversion to linked
data requires that a user 1) define a “skeleton” for the final RDF as a tree structure, 2)
maps table columns to nodes of the tree. Our tool combines these steps, providing a
simpler interface by keeping all of the interactions for RDF mapping situated with the
original data table. The users also retain the flexibility to define or change the schema
as they go along. RDF Refine also allows certain types of conversions through
"transposition" that requires the user to change the structure of the original tabular
data in order to perform the conversion. Our approach enables similar modifications
while maintaining the original structure of the CSV file.
Anzo Express2 by Cambridge Semantics enables semantic annotation of Excel
spreadsheets, and also provides various other functions, including an OWL ontology
editor. Our tool focuses on re-use of existing ontologies and community-standard
vocabularies. In contrast to both RDF Refine and Anzo Express, csv2rdf4lod has the
added functionality of generating provenance statements using popular vocabularies3
that enable tracking the original dataset through each of the stages from retrieval,
through conversion, and its publication as dump files or into a SPARQL endpoint.
3 Introducing the SemantEco Annotator
To illustrate the primary translation techniques available through the annotator, we
consider data from the Darrin Freshwater Institute that tracks the quality of lake water
in New York State’s Adirondack region. The techniques are 1) row and cell-based
translation; 2) implicit and explicit bundling; and 3) leveraging OWL ontology
classes, properties, and individuals to formulate new RDF statements. Each of these
techniques is demonstrated and discussed in detail in our video linked at
https://github.com/apseyed/SemantEcoAnnotator/wiki.
3.1 Row and Cell-Based Translation
A simple RDF translation of a table generates a triple for each non-empty cell value,
where the subject, predicate, and object URIs are formed from the row, column, and
cell value, respectively. The subject URI is either generated from a convention or
1
http://refine.deri.ie/
2
http://www.cambridgesemantics.com/products/anzo-express
3
http://prefix.cc/void,dcterms,prov
constructed from a column’s cell value. This default technique as row-based
translation, since every triple generated from a row shares the same subject.
In more complex tabular arrangements, each row may instead contain many
subjects that are described across multiple columns, and are related direct or
indirectly. The technique for this pattern is cell-based translation. Figure 1 illustrates
a scenario where it is useful for columns F and G, which contain measurements of
NH4+ and NO3, respectively, for a water sample. Direct or indirect features of the
measurements are contained in other cells in the same row, and relate back to the
sample. Columns are designated for cell-based translation via a context menu.
Translating the indirect relationships requires bundling, which we discuss in the next
section.
Fig. 1. An excerpt of water quality data from Darrin Freshwater Institute including information
about the sample (Cols. A-E) and two of its measurements (Cols. F,G). Directed arcs above the
table illustrate relationships between what is described in columns, which our annotator helps a
user describe and ultimately generates for the user via translation into RDF.
3.2 Bundling
Sometimes a column represents a description of an entity identifiable in another
column. The Date (col. B) describes when the Sample (col. A) was taken; the user can
capture this relationship by creating an explicit bundle. In contrast, columns may
describe an unspecified, or implicit set of entity not captured by another. For the
water sample data, the Lake Name (col. C), the depth to the bottom of the lake (col.
D, Z Max), and the depth of sample collection (col. E, Sample Z) together represent a
sample location. The user can represent the implicit entity by bundling the columns
that represent it together, in an implicit bundle. Bundled columns are visually “pushed
down” into a new row in the header to show they are now grouped, and subordinate
aspects of the entity they are bundled into. After the bundling selection is complete,
the user has a choice of one of the existing column identifiers for explicit from a drop-
down list, or declaring it an implicit bundle.
For the water measurement columns (F, G) designated for cell-based conversion,
there is vital unit information in the header. We provide a subject annotation feature
to directly assert out-of-band triples about the measurement units. Annotations are
created with the context menu and accept dragged properties and classes to generate
further triples. We illustrate these steps in our demonstration video.
3.3 Enhancing RDF Translation with OWL Ontologies
In connection with the aforementioned translation techniques, we provide
mechanisms for using OWL ontologies. The right side of the interface includes a
catalog of ontologies to choose from, which in turn populates trees with choices of the
ontologies’ classes, properties, and datatypes, for adding semantics to the resultant
RDF data. If an object, data, or annotation property node is dragged from the tree
interface into a column, the triples generated for that column will use the given
predicate. Dragging a class node into the column will type the object of the triple as
an instance of the given class. We also employ restrictions to selections based on
previous choices for a column. This is highlighted in our demonstration video,
through use of various scientific-observation ontologies such as OBO-E4.
4 Discussion and Future Work
Once enhancements are committed RDF files for the enhancement parameters and the
RDF produced by csv2rdf4lod are downloadable for use as LOD. In the future we will
leverage logical restrictions from ontologies to guide the user to constructs that are
most appropriate for subsequent enhancements. We also plan to extend the ontology
menu to load any ontology listed in catalogs such as BioPortal3, as well as via URI.
While the annotator enables translation of data into RDF using existing ontologies
and vocabularies, other components can enable mappings of terms within data
packages to ontology concepts. This environment would essentially enable generation
of semantically enriched data as linked data to enable better search capabilities.
References
1. Lebo, T., Erickson, J.S., Ding, L., Graves, A., Williams, G.T., DiFranzo, D., Li, X.,
Michaelis, J., Zheng, J., Flores, J., Shangguan, Z., McGuinness, D.L., and Hendler, J.
Producing and Using Linked Open Government Data in the TWC LOGD Portal. In Linking
Government Data, pages 51–72. New York, 2011. 10.1007/978-1-4614-1767-5_3
2. Seyed, A. P., Lebo, T., Patton E., McCusker J., and McGuinness D. L. SemantEco: A Next-
Generation Web Observatory. 1st International Web Observatory Workshop. 22nd
International World Wide Web Conference, Rio de Janeiro, May 13-17, 2013.
3
https://semtools.ecoinformatics.org/oboe
4
http://bioportal.bioontology.org/ontologies