\section{Semtools Framework} \label{sec:framework} The Semtools project has focused development efforts on three main components: a Java library to access and manipulate OBOE ontologies and semantic annotations, an annotation plugin for the Morpho metadata editor, and query extensions for the Metacat data catalog. Below we briefly describe the OBOE model, the semantic annotation approach used by Semtools, and the corresponding software components. For a more in-depth presentation of OBOE see \cite{madin07:_ontol_for_descr_and_synth,bowers08}. \begin{figure}[!t] \centering \includegraphics[width=0.494\textwidth]{images/oboe} \caption{Main classes and properties of the extensible observation ontology (OBOE). While shown using the Unified Modeling Language (UML), the model is defined as an OWL-DL ontology.} \label{fig:oboe} \vspace{-12pt} \end{figure} \subsection{The OBOE Observational Model} \figref{fig:oboe} shows the main modeling constructs of OBOE (see: \url{http://ecoinformatics.org/oboe/oboe.1.0/oboe-core.owl}). An {\em observation} is made of an {\em entity} (e.g., biological organisms, geographic locations, environmental features) and serves to group a set of measurements together to form a single ``\emph{observation event}''. A \emph{measurement} assigns a value to a {\em characteristic} of the observed entity (e.g., the weight of a plant), and can also include \emph{standards} (e.g., units as well as standards for coded values) and collection \emph{protocols}. An observation can occur within the surrounding \emph{context} of other observations (e.g., as part of a temporal or spatial context), and context may include a named relationship (e.g., ``partOf'', ``within'') that existed during the observation event. A key feature of OBOE is that it allows properties (characteristics and relationships) of entities to be asserted without being interpreted as \emph{inherently} (i.e., {always}) true of the entity. Depending on the context in which the entity was observed or how the measurements were performed, an entity's properties may take on different values. OBOE allows RDF-style assertions about entities to be contextualized, and thus different values can be assigned for the same entity under distinct contexts, which is a crucial feature for modeling ecological as well as many other types of scientific data \cite{bowers08,mungall07:_repres_phenot_in_owl}. In addition, OBOE is currently implemented as an OWL-DL ontology that can be easily used with (or extended by) other ontologies for specifying domain-specific types of entities, characteristics, measurement standards, protocols, and relationships. For instance, the Semtools project has defined specific OBOE extensions in collaboration with the Santa Barbara Coastal Long-Term Ecological Research Project as well as through ongoing collaborations with other projects, and general extensions exist for OBOE that define a number of common entities, measurement units, and corresponding physical characteristics. \begin{figure}[!b] \centering \includegraphics[width=0.5\textwidth]{images/kelp-mass-model.pdf} \caption{Partial OBOE semantic annotation for Kelp sampling data. Shaded nodes represent ontological concepts; rectangular nodes are data table attributes mapped to OBOE measurement characteristics.} \label{fig:kelp-mass-model} \end{figure} \subsection{Semantic Annotations} A semantic annotation consists of two parts: (i) a ``configuration'' of the observation model containing the specific entities, characteristics, observations, measurements, and so on (drawn from one or more domain ontologies) that appropriately capture the semantics of the data set; and (ii) a mapping between the attributes in the data set to specific measurements defined in the model configuration. \figref{fig:kelp-mass-model} shows a high-level example of an annotation defined for a simple Kelp sampling data set. Here, the data set consists of five attributes (bottom of \figref{fig:kelp-mass-model}). Each attribute is mapped to a specific measurement type (where only the characteristic of each measurement type is shown), and measurement types are organized into observations of specific Kelp entities (shown of type ``Macrocystis''), temporal points (denoted by date-times), and spatial locations (given as site names). Each measurement associated with a Kelp observation is assumed to have occurred within the site and during the given time as specified by the context relationships. \begin{figure*}[!t] \centering \includegraphics[width=1.0\textwidth]{images/morpho-annotation-widget.png} \caption{Morpho metadata editor with Semantic plugin. The fill-in-the-blank interface uses natural language descriptions for intuitive editing. A searchable, hierarchical browser is used to select concepts from domain-specific ontologies.} \label{fig:morpho-annotation} \end{figure*} Semantic annotations can be used to facilitate discovery and integration of heterogeneous data sets. For instance, combining semantic annotations with OBOE, it is possible to discover data sets based on searches expressed over types of observations and measurements of interest. As simple examples, users can pose queries such as ``\emph{find all data sets containing observations of Kelp}'' and ``\emph{find all data sets containing Mass measurements of Kelp}''. Both of these queries would return the example data set in \figref{fig:kelp-mass-model} since the attribute {\tt WET} is linked to a WetMass measurement for observations of Macrocystis (where WetMass is defined as a special kind, or \emph{subclass} of Mass, and Macrocystis is defined as a subclass of Kelp). Using semantic annotations in this way can help to increase both query recall and precision over standard keyword-based approaches \cite{berkley09:_improv_data_discov_for_metad}. In particular, by defining terms as subclasses of other terms (e.g., Macrocystis as a subclass of Kelp), term expansion can be used to increase the number (recall) of data sets returned (where subclasses of query terms are also searched). The precision of the result can be improved since queries may specify the desired connections between terms (e.g., measurements of Mass for Kelp observations) as opposed to returning all data sets that simply mention the terms but without any explicit connections (i.e., where Mass was measured, but not for Kelp samples). % Queries can additionally include context constraints, e.g., Mass % measurements of Kelp within a specific spatial or temporal context. Annotations also help facilitate integration by allowing tools to align data set attributes based on their declared measurement and observation types. In general, semantic annotations provide a formal description of attribute semantics, whereas in many commonly-used metadata formats for describing data sets, only informal text-based descriptions of data attributes are permitted. In the case of EML, there is some overlap between these two mechanisms---particularly with respect to measurement standards---however, semantic annotations extend this approach by providing a general mechanism to formally associate concepts drawn from domain ontologies to attributes. In the Semtools framework, we employ an XML serialization syntax for semantic annotations that is compatible with EML but that is stored separately from the EML documentation of a data set (allowing, e.g., annotations to be used independently of EML or with other metadata standards if needed). In addition, semantic annotations can be used to ``materialize'' a given data set into a set of triples conforming to the model configuration given in the annotation. In other words, a tabular data set such as the one shown in \figref{fig:kelp-mass-model} can automatically be converted into a corresponding collection of observation and measurement instances. This in turn enables a simple form of structural integration, where instead of having a large number of different tabular data structures, all data is represented using the standard set of structures defined by the OBOE model (see \figref{fig:oboe}). Thus, materializing a data set in this way provides a more uniform structural representation that can make a number of discovery and integration tasks easier. For instance, materialization can be used to increase query expressivity by allowing searches of the form ``\emph{find all data sets containing Mass measurements of Kelp with values less than or equal to 5 grams}'', which in our example can be answered by generating (i.e., materializing) the measurements associated with the {\tt WET} and {\tt DRY} attributes in the data set of \figref{fig:kelp-mass-model}. % Because the Annotation represents a potentially subjective % perspective on the data, they are stored independently as XML; the % former referencing the latter. The Annotation structure largely % mirrors the core classes in OBOE. Observations are composed of % Measurements of a specific Entity and are represented by a % collection of Characteristics (usually just one) collected using a % defined Protocol and Standard (i.e. unit). Tablular data object % attributes are mapped to a Measurement such that a collection of % attributes usually pertain to the Entity of a shared Observation % (\figref{fig:kelp-mass-model}). These Observtions can provide % context for other Observations such that the structure of the % observational data model and collection paradigm are formally % represented in the Annotation. \subsection{The Semantic Mediation API} The Semantic Mediation API includes basic ontology management features, annotation manipulation capabilities, and simple concept navigation and visualization components. The API is intended to be a centralized toolkit for use in multiple application contexts (on either client or server deployments). The Semantic Mediation API uses both the OWL API \cite{owlapi} for ontology management services including ontology parsing, serialization, and simple class and property exploration as well as the Pellet description-logic reasoner \cite{pellet} for classification and exposing inferred axioms in source ontologies. The inference services exposed through the Semantic Mediation API are used in both discovery and integration features described below. % is particularly powerful when utilizing OBOE Measurement 'templates' % that express exactly what (Entity and Characteristic) can be % observed and how (Standard and Protocol). In our current Morpho and Metacat extensions, semantic annotations are managed and stored automatically in an underlying, local relational database. While it is also possible to use in-memory approaches for storing and querying annotations, we found the overhead to be prohibitive when large numbers of data sets are managed. % store and query annotations inmemory % approachesfor storing for scalable persistence and improved query % performance, which we found to be essential when large numbers of data % sets are supported. , where the overhead of in-memory annotation % queries having proven to be prohibitively limiting for large sets of % data. \subsection{The Morpho Editor Plugin} The semantic-annotation editor plugin for Morpho provides a front-end to the Semantic Mediation API and allows data owners and curators to define annotations for existing EML data descriptions. The editor provides a simple ``fill-in-the-blank'' style form-based interface with a searchable hierarchical concept selection widget (see \figref{fig:morpho-annotation}). The plugin seamlessly integrates with a standard Morpho installation and provides semantic query capabilities for locating data packages, marking up data sets within a package using semantic annotations, and saving annotations locally or to a shared repository where they can be discovered and explored by other users. The annotation editor in Morpho allows a user to view the data set being annotated as they fill in (by selecting an appropriate ontology term) the characteristic, measurement standard, protocol, and associated entity for each data set attribute. Users can also specify whether an observation spans multiple columns, and can provide context relationships between attributes (i.e., observations). The editor provides a number of additional features including the ability to view the entire annotation (similar to \figref{fig:kelp-mass-model}) and to specify additional mapping constraints for observations and measurements. \subsection{Metacat Query Extensions} The semantic plugin for Metacat augments Metacat's existing metadata storage and search by allowing annotations to be saved and queried alongside the metadata and data that they annotate. In addition to traditional keyword and spatial search criteria, the Metacat plugin allows semantic criteria to be included where they may either increase query recall using term-expansion (i.e., traversing the class subsumption hierarchy) or refine the result set by limiting matches to data sets that contain the specified observational model (e.g., combinations of OBOE-compatible entity, characteristic, measurement standard, or protocol concepts). The observational model can be leveraged further by materializing the annotation and data artifact (via the Data Manager Library \cite{leinfelder10:_metad_driven_approac_to_loadin}) into a fully instantiated OBOE model and inspecting (and querying over) the observational values themselves. %%% Local Variables: %%% mode: latex %%% TeX-master: "main" %%% End: