\begin{figure*}[!t]
\centering
\includegraphics[width=1.0\textwidth]{images/metacat-query.png}
\caption{Semantic data query web interface. Data packages containing observations of Kelp Wet Mass less than or equal to 5 [grams] are returned. Additional search options and compound query criteria can be specified within the other tabs. Matches can be saved in the data cart for later exploration.}
\label{fig:metacat-query}
\end{figure*}


\section{Discovery and Integration}
\label{sec:application}

In this section we describe the new data discovery and integration
applications we have built using the components described above as
part of the Semtools project.

\subsection{Concept Query} 

The semantic query interface (see \figref{fig:metacat-query}) is
implemented as a Web application over Metacat that primarily supports
locating data sets by how well their observational models match the
given criteria. The interface provides \emph{structured} as opposed to
\emph{unstructured}, i.e., keyword-based queries. In particular, query
criteria given by users largely mirror the structure of a semantic
annotation in that combinations of Entity, Characteristic, and
Protocol are specified and optionally compounded when increased
precision is sought.

As discussed above, by leveraging the relationships defined and
inferred from the ontology we are able to increase recall beyond what
is possible for simple keyword-based searches
\cite{berkley09:_improv_data_discov_for_metad}. Broad queries return
direct matches as well as subclass matches. The queries can be quickly
refined when using the Web application by allowing rapid exploration
of the data repository without having to define complete observational
queries \emph{de novo}. The interface allows users to specify
individual classes of a measurement as well as pre-configured
measurement types (representing standard data set attribute types) as
defined in OBOE compatible ontologies to enable a single concept to
proxy its constituent parts, namely the characteristics of particular
entities that can be measured with a set of protocols and
standards. This short-hand query generation can save users time in
specifying their queries, and highlights a compelling reason for using
OBOE extension ontologies. Measurement templates can also be leveraged when creating or
editing semantic annotations in the Morpho interface.

Using compound semantic query criteria applies a finer-grained filter
on the data sets that are returned. Results can be restricted to only
those data sets that include measurements for a set of specific
characteristics of a particular observational entity. Furthermore, a
query can specify that those measurements come from precisely the
\emph{same instance of that entity}; a feature that fully exercises
the comprehensive observational structure expressed in the annotation
and enables higher query precision as described above.

\subsection{Data-Level Query} 

For even more precise recall, the OBOE model can be (partially)
materialized (see above) during the query stage after which a data
range filter can be applied. Different techniques are available for
merging the annotation with the data that it describes, but for
performance reasons a hybrid approach has been adopted in which
preliminary search results from a structured query are collated and
only that subset is materialized. Because our corpus is described
using EML in conjunction with the annotation syntax, the Data Manager
Library \cite{leinfelder10:_metad_driven_approac_to_loadin} is used to
load the described raw data (into a relational database) while the
annotation informs the correct use of the Data Manager query and
filtering features. For any measurements that match the concept query
criteria, we verify that those measurements (e.g., attributes) contain
data values within the range specified in the initial semantic data
query and return the data packages that contain them (see
\figref{fig:metacat-query}).

\subsection{Data Integration} 

The materialization routine for semantic data queries can help in
enabling data integration. In addition to inspecting data for values
within a range and returning the data sets that contain a match, the
data integration feature of the Semtools Web application goes further
by constructing a synthetic data product (table) that represents the
complete results of the query in terms of both the attributes and the
values that are returned. Each original data set may have very
different syntactic structures (e.g., column number, naming, order)
but could share a subset of attributes that are semantically
compatible as defined in accompanying annotations. These compatible
attributes can then become the basis for a synthetic data set.
\figref{fig:integration} illustrates the data integration support
provided in the current implementation of the Web
application. Consider the two data packages (denoted A and B) in
\figref{fig:integration}.  Annotations (denoted C and D) are used to
determine semantically equivalent data attributes contained in the
data sets (denoted by E and F). The attributes {\tt plot} and {\tt
  site} are considered equivalent measurements of the characteristic
Location; attributes {\tt weight} and {\tt wt} both map to the same
characteristic Mass. The Semantic Mediation API computes an
equivalence among attributes based on their corresponding
annotations. The Data Manager Library is then used to load the data
sets and then query each data set to produce and merge a synthetic
result data set.

While this approach provides a preliminary form of data-level
integration, we are currently developing additional algorithms for
determining compatibility of annotated measurements (e.g., to include
unit information such as that gram and ounce are both mass units) and
for converting measurement values using ontologically-defined unit
conversions (e.g., 1000 milligrams in a gram), which will further
support automated data integration through the Web application.


\begin{figure}
\centering
\includegraphics[width=0.45\textwidth]{images/integration.pdf}
\caption{Integration query across multiple data packages (A, B).
  Annotations (C, D) determine semantically equivalent data attributes
  contained in the data objects (E, F). Attributes {\tt plot} and {\tt
    site} are considered equivalent measurements of the characteristic
  Location; attributes {\tt weight} and {\tt wt} both map to the same
  characteristic Mass. The Semantic Mediation API utilizes the Data
  Manager Library to load and query the source data informed by
  semantic similarities between the structurally disparate data
  attributes.}
\label{fig:integration}
\end{figure}

%%% Local Variables: 
%%% mode: latex
%%% TeX-master: "main"
%%% End: