=Paper=
{{Paper
|id=Vol-1442/paper_28
|storemode=property
|title=Annotation-Based Method for Linking Local and Global Knowledge Graphs
|pdfUrl=https://ceur-ws.org/Vol-1442/paper_28.pdf
|volume=Vol-1442
|dblpUrl=https://dblp.org/rec/conf/ontobras/CavotoS15
}}
==Annotation-Based Method for Linking Local and Global Knowledge Graphs==
https://ceur-ws.org/Vol-1442/paper_28.pdf
Annotation-Based Method for Linking Local and Global
Knowledge Graphs
Patrícia Cavoto1, André Santanchè1
1
Laboratory of Information Systems (LIS)
Institute of Computing (IC)
Univeristy of Campinas (UNICAMP)
Campinas – SP – Brazil
patricia.cavoto@gmail.com, santanche@ic.unicamp.br
Abstract. In the last years, the use of data available in “global graphs” as
Linked Open Data and Ontologies are increasing faster and bringing with
them the popularization of the graph structure to represent information
networks. One challenge, in this context, is how to link local and global
knowledge graphs. This paper presents an approach to address this problem
through an annotation-based method to link a local graph database to global
graphs. Different from related work, the local graph is not derived from a
static dataset, but it is a dynamic graph database evolving along the time,
containing connections (annotations) with global graphs that must stay
consistent during its evolution. We applied this method over a dataset with
more than 44,500 nodes, annotating them with the values found in DBpedia
and GeoNames. The proposed method is an extension of our ReGraph1
framework that bridges relational and graph databases, keeping both
integrated, synchronized and in their native representations, with minimal
impact in the current infrastructure.
1. Introduction
Real-world phenomena as biological processes, social networks and information systems
have been increasingly modeled as networks, where nodes can represent individuals,
computers, species, proteins, etc. and links the interaction among them. Recent research are
pointing graphs as the fitted structure to store this kind of data, in which the relations
among data elements are as important as the elements themselves. In the biology field,
there are many uses for graphs, including metabolic networks, chemical structures and
genetic maps [Vicknair et al. 2010]. The challenge is how to explore the network "behind"
data available in existing information systems for analysis when data is stored in formats
that do not valorize such network structure.
This challenge motivated our proposition of ReGraph, a framework inspired in the
OLAP approach, which creates a special local graph database designed for network-driven
analyses, aligned with an existing relational database. We applied ReGraph to taxonomic
data from FishBase2 to create FishGraph [Cavoto et al. 2015].
1
http://patricia.cavoto.com.br/regraph/
2
http://www.fishbase.org/
In this paper, we present an automatic annotation-based method to link our local
graph database to global graphs from the Semantic Web, applied to link FishGraph data
with DBpedia. Our method contributes in the data quality analysis, in the enrichment of the
local database and in building the Giant Global Graph.
This is a work in progress concerning how to relate data from a local graph, stored
in a graph database, with global graphs. Different from related work, our local data
repository is not a static set of documents or tags to be enriched, but a dynamic graph
database. It annotated content evolve along the time, bringing challenges, addressed in this
research, of how to manage this hybrid graph (local and global) maintaining its consistency
during the evolution.
The remainder of the paper is organized as follows. Section 2 presents the related
work. Section 3 details our ReGraph framework. Section 4 presents our annotation-based
approach to enrich data using ontologies. Section 5 presents our conclusions and future
work.
2. Related Work
There are several contexts in which annotations are related to the Semantic Web resources
(LOD and ontologies). The annotations are produced manually, semi-automatically or
automatically, helping the improvement of information retrieval, knowledge reuse and
information exchange [Oren et al. 2006]. There are works proposing annotations over wiki
pages [Oren et al. 2006] and publishing personal notes as linked data in semantic blogs
[Drǎgan et al. 2010].
Several initiatives focus in how to reach semantic concepts to relate them to
resources. In a survey of semantic search approaches, the authors present an overview and
a classification of the existing methods for searching and browsing linked data and
ontologies [Mangold 2007]. In [Alm et al. 2014] the authors propose a model to extract
characteristic features from semantic annotations by importing the ontology concepts and
their taxonomic relationships. Another work uses taxonomic distance measures to compute
relatedness of the ontological annotations [Palma et al. 2014].
The work presented in [Santos et al. 2011] propose an architecture to discover
information sources through the use of semantic search techniques in a corporative
metadata repository. The process begins with an initial keyword list, followed by the query
reformulation process that expands this list, adding semantically related terms and creating
a new query to run on semantic annotations.
In [Amanqui et al. 2013], the authors developed a semantic search application that
uses semantic web key concepts for information retrieval. They have proposed an
architecture for semantic search that maps concepts of the OntoBio domain ontology to a
database from the National Institute for Amazonian Research (INPA), which has
collections of insects, fishes, and mammals, totalizing over 16,500 species.
As mentioned before, this work differs since it introduces a graph database
perspective over the locally annotated data, which dynamically evolve along the time and
must stay consistent.
3. ReGraph
As mentioned before, this method is an extension feature in our ReGraph framework,
which provides a bridge integrating relational and graph databases, keeping both
synchronized in their native representations. In this section, we briefly explain how the
ReGraph framework works and the data conversion process from a relational to a property
graph database.
3.1. The ReGraph Framework
The FishBase data is stored in a relational database. Besides the existing relational
database, ReGraph produces a parallel property graph database (FishGraph), to perform
network analyses and to link data with Semantic Web.
Starting from a relational database, ReGraph allows mapping its data into a
property graph database, generating a mapped subgraph. It is also possible to further
create manual and automatic annotations over this data, generating an annotation
subgraph. Both subgraphs, mapped and annotation, are connected in the graph database.
ReGraph keeps relational and graph databases in their native forms and has a synchronism
module that reflects in the graph database changes executed in the relational database. The
graph database is focused in the analysis on the relations among data elements.
3.2. From FishBase to FishGraph using the ReGraph framework
As previously mentioned, FishGraph concerns an application of ReGraph in the FishBase
information system. We have mapped the taxonomic classification of fishes from FishBase
to FishGraph - see details in [Cavoto et al. 2015]. The taxonomic classification of a species
includes: Kingdom, Phylum, Class, Order, Family, Genus and Species. As FishBase has
only species of fishes, it does not register Kingdom and Phylum, once that all fishes belong
to the same Kingdom and Phylum. This data was compared to the taxonomic classification
defined in DBpedia, generating a comparison annotation type.
In order to generate a new annotation type, we have selected also the table Country,
representing countries where species are found. Figure 1 shows the graph model for the
taxonomic classification and country data generated in the graph database, in which we
have nodes and, associated with them, their respective properties and edges connecting it to
each other.
Figure 1 - Graph Model for Taxonomic Classification and Countries
We used the country information in the graph database to link them with
GeoNames, a geographical knowledge base that covers all countries and contains over
eight million placenames. Data retrieved from GeoNames generated new nodes and edges
in the graph database, enriching it and bringing more details to the performed analyses.
After the migration of the related data, we generated in the graph database 226,284 edges
and 44,701 nodes, in which we have: 311 countries; 32,957 species; 10,790 genera; 572
families; 65 orders and 6 classes.
4. Automatic Annotation-Based Method
Annotations can improve the understanding and the quality of the data adding extra
information. We propose a method that allows creating automatic annotations over the
existent data in a property graph database. These annotations will be created through a
direct connection with existing ontologies and LOD, available on the Web, e.g.,
GeneOntology, GeoNames and DBpedia. In this section, we detail our automatic
annotation-based method and the two distinct annotation types implemented: Comparison
and New. Independently of the annotation type, local data is related to Web data through a
match function that compares strings to find the proper resource.
A distinctive feature of our approach is to differentiate the annotation subgraph
(produced here) from the mapped subgraph (mapped from the relational database). The
mapped subgraph cannot be directly changed in the graph database, since it is the product
of a one-way synchronization originated in the relational database. Synchronization rules
avoid updates in the mapped subgraph that will create inconsistencies with the annotation
subgraph.
4.1. The Comparison Annotation Type
The main goal in the Comparison annotation type is to record comparisons of data stored in
the local graph database with third party sources available on the Web. To execute this
type of automatic annotation, it is necessary to define the "subject query" that will return
the data from the property graph database that will be subject of the comparison.
The order of the data returned by the subject query is determinant to the correct
execution of the process: (i) the first value will be the identifier of the node, helping the
annotation process; (ii) the second value will be the key matched with the ontology
identifiers; it will be used by the match function to retrieve data on the Web; (iii) for each
of the remaining values, it is necessary specify the direct path in the ontology to reach it,
linking the returned values with the specific value in the ontology; it is possible to define
two paths in the ontology for each value returned by the subject query.
The result of this comparison will produce an annotation over the first node
returned by the subject query. This annotation is added in the graph database as a property
of the node, in which there are three possible values, annotated automatically:
- Equal: indicate elements that have the same value in the graph database and in the
external ontology. This kind of annotation can improve the quality and the confidence of
the data, through a double check validation.
- Not Found: represent existing elements in the graph database that was not found in the
referred ontology. It can indicate: data in the graph database has spelling mistakes; the
specified data does not exist in the referred ontology; data was updated in one of the
sources, and was not in the other; etc.
- Divergent: represent data that have a divergence compared to the referred ontology. In
can indicate: incorrect data in the graph database or in the ontology. This value is
defined as a recommendation to review data. In addition, a new node is added, linked
with the existing node, containing the exact data in the ontology for traceability.
4.2. The New Annotation Type
In the New annotation type, we produce new nodes, edges and/or properties, to improve the
analysis and results. In this annotation type, it is necessary to specify in the "subject query"
only two values: (i) the first one will be the identifier of the node, helping in the annotation
process; (ii) the second one represents the key in the graph database matched with the
respective identifier of a resource in the ontology; it is used by the match function to
retrieve data on the Web. The second step is to define the ontology path to search.
Both data are the starting point to search in the ontology. For each information to
be retrieved from the ontology and inserted in the graph database it is necessary specify: (i)
ontology information: direct path in the ontology to retrieve the required information; (ii)
annotation creation: how the annotation will be created in the graph database: as a node or
property. The new node will be connected with the existing node by an edge that has its
label also defined. In the property option, a defined property will be created over the
existing node. In both cases, the value of the property will be the value found in the
specified ontology.
5. Conclusions and Future Work
In this paper, we presented an automatic annotation-based method using ontologies, as an
extension of our project ReGraph that connects a relational database with a property graph
database, keeping both integrated, synchronized and in their native forms. It stands out for
its flexibility in defining the ontologies and values that will be retrieved, compared and
created, offering several possibilities to validate and enrich the graph database. Our method
contrasts with the related work since it introduces a graph database perspective over the
annotation-based connection between the local and global graphs. Annotations in the
annotated subgraph stay consistent with the existing mapped subgraph, even after its
evolution along the time.
We developed two distinct experiments to validate each proposed annotation type:
Comparison and New. In the Comparison experiment, we compared almost 33,000 species
of fishes from FishBase to validate their taxonomic classification with DBpedia. In the
New experiment, we used the 249 countries in the graph database to retrieve their continent
and information of GeonNameID and population from GeoNames.
Future work includes extending the functionality of ReGraph to allow retrieving
data from other web formats and to save the link to the resource in the graph database as
well as the "subject query" that generated it, helping in future repeated analysis and to
track provenance.
Acknowledgments
Research partially funded by projects NAVSCALES (FAPESP 2011/52070-7), the Center
for Computational Engineering and Sciences (FAPESP CEPID 2013/08293-7), CNPq-
FAPESP INCT in eScience (FAPESP 2011/50761-2), INCT in Web Science (CNPq
557.128/2009-9) and individual grants from CNPq and CAPES. Thanks to FishBase.org,
which provided the data used in this work.
References
Alm, R., Waltemath, D., Wolkenauer, O. and Henkel, R. (2014). Annotation-Based
Feature Extraction from Sets of SBML Models. In: Data Integration in the Life
Sciences 10th International Conference, DILS 2014, p. 81–95.
Amanqui, F. K., Serique, K. J., Lamping, F., et al. (2013). Semantic Search Architecture
for Retrieving Information in Biodiversity Repositories. In: VI Seminar on Ontology
Research in Brazil, Ontobras 2013, p. 83–93.
Berners-Lee, T. (2007). Giant global graph. Decentralized Information Group.
Bizer, C., Heath, T. and Berners-Lee, T. (2009). Linked data-the story so far.
International journal on Semantic Web and Information Systems, v. 5, p. 1–22.
Castilho, F. M. B. M., Granada, R. L., Vieira, R., Sander, T. and Rao, P. (2011). Ontology
enrichment based on the mapping of knowledge resources for data privacy management.
In: CEUR Workshop Proceedings, v. 776, p. 85–96.
Cavoto, P., Cardoso, V., Vignes Lebbe, R. and Santanchè, A. (2015). FishGraph: A
Network-Driven Data Analysis. In: 11th IEEE International Conference on e-Science
2015, p.1-10.
Drǎgan, L., Passant, A., Handschuh, S. and Groza, T. (2010). Publishing semantic
personal notes as linked data. In: CEUR Workshop Proceedings, v. 674, p. 1–2.
FishBase Consortium (2015). FishBase. http://www.fishbase.org, July, 2015.
Mangold, Christoph. (2007). A survey and classification of semantic search approaches.
International Journal of Metadata, Semantics and Ontologies, v. 2, n. 1, p. 1-23.
Oren, E., Delbru, R., Möller, K., Völkel, M. and Handschuh, S. (2006). Annotation and
navigation in semantic wikis?. In: CEUR Workshop Proceedings, v. 206, p. 58–73.
Oren, E., Möller, K. H., Scerri, S., Handschuh, S. and Sintek, M. (2006). What are
Semantic Annotations?. Technical Report. http://www.siegfried-
handschuh.net/pub/2006/whatissemannot2006.pdf, July, 2015.
Palma, G., Vidal, M.-E., Raschid, L. and Thor, A. (2014). Exploiting Semantics from
Ontologies and Shared Annotations to Partition Linked Data. In: Data Integration in
the Life Sciences 10th International Conference, DILS 2014, p. 120–127.
Santos, V. Dos, Baiao, F. A. and Tanaka, A. (2011). An architecture to support
information sources discovery through semantic search. In: 2011 IEEE International
Conference on Information Reuse & Integration, p. 276–282.
Vicknair, C., Macias, M., Zhao, Z., et al. (2010). A comparison of a graph database and a
relational database. In: ACM Southeast Regional Conference, pp. 1–6.