Download from Data Mining to Sign Management

Brigham Young University
BYU ScholarsArchive
International Congress on Environmental
Modelling and Software
6th International Congress on Environmental
Modelling and Software - Leipzig, Germany - July
2012
Jul 1st, 12:00 AM
Knowledge Discovery for Biodiversity: from Data
Mining to Sign Management
Noel Conruyt
David Grosser
Régine Vignes Lebbe
Follow this and additional works at: http://scholarsarchive.byu.edu/iemssconference
Conruyt, Noel; Grosser, David; and Lebbe, Régine Vignes, "Knowledge Discovery for Biodiversity: from Data Mining to Sign
Management" (2012). International Congress on Environmental Modelling and Software. 297.
http://scholarsarchive.byu.edu/iemssconference/2012/Stream-B/297
This Event is brought to you for free and open access by the Civil and Environmental Engineering at BYU ScholarsArchive. It has been accepted for
inclusion in International Congress on Environmental Modelling and Software by an authorized administrator of BYU ScholarsArchive. For more
information, please contact [email protected].
International Environmental Modelling and Software Society (iEMSs)
2012 International Congress on Environmental Modelling and Software
Managing Resources of a Limited Planet, Sixth Biennial Meeting, Leipzig, Germany
R. Seppelt, A.A. Voinov, S. Lange, D. Bankamp (Eds.)
http://www.iemss.org/society/index.php/iemss-2012-proceedings
Knowledge Discovery for Biodiversity:
from Data Mining to Sign Management
1
1
2
Noel Conruyt , David Grosser , Régine Vignes Lebbe
LIM - IREMIA – EA 25-25 – Laboratory of Mathematics and Computer Science
University of Reunion Island, 2 rue Joseph Wetzell, 97490, Sainte-Clotilde, France
2
LIS, UMR 7207-CR2P, Université Pierre et Marie Curie, Paris Universitas, France
[email protected]
1
Abstract: Knowledge discovery from data in environmental sciences is becoming
more and more important nowadays because of the deluge of information found in
databases of digital ecosystems, coming altogether from institutions and amateurs.
For example in biodiversity science, all these data need to be validated by
specialists with the help of Intelligent Environmental Decision Support Systems
(IEDSSs), then enhanced and certified into qualitative knowledge before reaching
their audience. Data mining through classification or clustering is the dedicated
inductive process of grouping descriptions based on similarity measures, then
building classes and naming them. Later, the formed concepts can be reused for
identification purpose with new observations. The problem is that when using such
knowledge-based systems, we tend to forget the fundamental role of subjects (endusers) in the definition, observation and description of objects. In order to get good
identification results, a consensus must be found between these experts and
amateurs for interpreting correctly the observed objects. Thus, a new method of
Knowledge discovery is necessary by switching from Data mining to Sign
management. The method focuses on the process of building knowledge by
sharing signs and significations (Semiotic Web), more than on knowledge
transmission with intelligent object representations (Semantic Web). Sign
management is the shift of paradigm for Biodiversity Informatics that we have
investigated in such domains as enhancing natural heritage with ICT. In this paper,
we will present Sign management and illustrate this concept with two knowledge
bases built in La Reunion Island for corals’ classification with IKBS (Iterative
Knowledge Base System) and plants identification with Xper2 software platform.
Keywords: Data mining; Sign management; Knowledge discovery; Environmental
sciences; Biodiversity informatics.
1
KNOWLEDGE DISCOVERY IN ENVIRONMENTAL SCIENCES
th
At the end of the 20 century, a lot of researches have been made in computer
science for enhancing the use of computers for data analysis (Breiman et al. 1984)
(Diday 1994), machine learning (Mitchell 1997) and knowledge management
(Skyrme 1998). Across a wide variety of fields such as science, business, industry,
data have been collected and accumulated at a dramatic pace, generating a new
st
challenge for the 21 century, called the Big Data problem (Nature 2008) (Cukier
2010). In the domain of environmental sciences, the management of natural
resources such as soil, water, air, biodiversity, alimentation is facing the same
deluge of information in databases (Fukuda 2009).
To tackle this problem, Fayyad et al. (1996) explained that there is an urgent need
for a new generation of computational theories and tools to assist humans in
extracting useful information or knowledge from the rapidly growing volumes of
Noel Conruyt et al. / Knowledge Discovery for Biodiversity: from Data Mining to Sign Management
digital data. These topics were the subject of the emerging field of Knowledge
Discovery in Databases (KDD). KDD is the “non-trivial process of identifying valid,
novel, potentially useful, and ultimately understandable patterns in data”.
st
But in the digital era of the 21 century, with these torrents of new data on the
Web, discovering knowledge in environmental sciences is adding a more complex
challenge (Poch et al. 2004). Complexity comes from the fact that environmental
1
data are natural (i.e. not invented by humans). They are polymorph (variable),
incomplete (missing), noisy (difficult to observe), and therefore source of errors
(Gibert et al. 2008). This is particularly true for data management in the new field of
Biodiversity Informatics (Sarkar 2009). It leads to different interpretations of data by
2
people, called information . These pieces of information can be managed to
3
discover knowledge but information issued from human sensors (sensitive organs
such as the eye) are more subjective than data issued from machine sensors
(probes).
Thus, Knowledge Discovery in Biodiversity Informatics (KDBI) is not only a problem
of extracting knowledge from a flow of data with rational mining techniques and
algorithms (Gibert 2010). It is also a problem of human and machine interaction
with qualitative contents. Quality is a matter of good interpretation, visualisation
and signification of the triptych (data, information and knowledge) from a subject
viewpoint. Indeed, the challenge of KDBI is to make sense of the deluge of
contents and forms of delivery altogether in space (geo location of data), in time
(cycling or iterative process of generating knowledge) and at a certain level of
observation (multi-layered study’s scale).
In fact, KDBI is embedded in the Life Science problem of understanding emerging
theories (Umwelt, Activity and Semiotics) and tools that are the foundations of the
concept of Sign management (Uexküll 1926) (Engeström 1987) (Peirce 1965). For
taking the right decision (i.e. make a diagnosis or identify a problem) and act
properly, the application of computer-assisted tools to extract knowledge from
experts in knowledge bases (by representation and classification of objects) is not
enough; Ontology building and case-based reasoning with the description logics
formats (OWL, RDFs, SPARQL) of the Semantic Web brings syntactical
advantages, but this is not sufficient for its utility and usability. We do need to know
the purpose of knowledge discovery and formulate the right question before
acquiring data. The alternative that we suggest for specimens’ identification is then
to apply a more natural or human-centred approach to knowledge discovery by
applying descriptive logics of the Semiotic Web, based on the aim to tackle real
usage. Our idea is to manage not only objective knowledge with formal
representations for ontologies (written notations). We must also share subjective
know-how or interpretations of knowledge with living significations (multimedia
annotations) that can be shown by individual subjects. As suggested by Frey
(2008), let us reinvent a more living knowledge discovery on the Web for
st
biodiversity management in the 21 century!
2
BIODIVERSITY INFORMATICS
Biodiversity Informatics is a young and rapidly growing field that brings information
science and technologies to bear on the data and information generated by the
study of organisms, their genes, and their interactions. In doing so, it is creating
unprecedented global access to information on biological species and their role in
nature (e-Biosphere 2009). The term "Biodiversity Informatics" is generally used in
the broad sense of applying to computerized handling of any biodiversity
information and must not be confused with the term “Bioinformatics” that
1
A Datum is a raw element of Information that is attached to a thing or entity (Object) and
qualifies it with a Value measured by an instrument.
2
Information is the interpretation of a Datum measured on an Object by a Subject. It is the
shaping of the datum in a message that circulates between Subjects by a channel. The
interpreter makes a more or less explicit characterization of the Datum with an Attribute
(character or variable) to qualify the Object.
3
Knowledge is the result of negotiating pieces of information from different Subjects by
analyzing and synthesizing them (through comparison, discrimination and generalization).
Noel Conruyt et al. / Knowledge Discovery for Biodiversity: from Data Mining to Sign Management
computerizes data in the specialized area of molecular biology. At University of
Reunion Island, our application domain in environmental sciences has been to
manage Biodiversity with Information and Communication Technologies (ICT) at
the level of Systematics. It is the scientific discipline that deals with listing,
describing, naming, classifying and identifying living beings (Matile et al. 1987)
(Winston 1999). In this domain, specific IEDSSs are needed to help taxonomists
describe data for classification and identification purposes (Conruyt and Grosser
2007) (Ung et al. 2010). This phenetic domain of knowledge discovery, also called
alpha e-taxonomy or cyber-taxonomy (Mayo et al. 2008) is concerned with
ontogenetic descriptions of specimens and taxa and is not to confuse with
phylogeny that deals with species history. Indeed, being able to describe, classify
and identify a specimen from morphological characters is a first step for monitoring
biodiversity, because it gives access to information relative to its species name
(Biology, Geography, Ecology, Taxonomy, Bibliography, Photography, etc.). These
tasks can be assisted in Biodiversity Informatics with a databases management
system for storing unstructured information, and a knowledge bases management
system for acquiring codified data and knowledge (Conruyt et al. 2008). These
modules are part of a Biodiversity Information System that we want to open to
4
5
other platforms such as GBIF and ViBRANT e-infrastructures by using Web
Services (Conruyt et al. 2010).
3
METHOD OF KNOWLEDGE DISCOVERY
In Grosser et al. (2010), we present an inductive approach for identifying specimen
descriptions with domain knowledge. An expert is assisted with computer science
decision support tools for helping him to define a descriptive model (ontology),
describe related cases, classify these descriptions and let other biologists identify
specimens. For these complex tasks, classical discrimination methods developed
in the frame of data analysis or machine learning, such as classification (Breiman
1984) or decision trees (Quinlan 1993) or others methods developed in the data
mining field such as association rules mining (Piatetsky-Shapiro 1991), Multifactor
dimensionality reduction (Zhu and Davidson 2007) or in computer vision (Mansur et
al. 2006) are not sufficient. They do not cope with relations between attributes, or
with missing data, and they are not also very tolerant to errors in descriptions. The
considered problem that we are faced with is to determine the class of a structured
description that is partially answered and eventually contains errors, from a
referenced case base, this last one being a priori classified by qualified experts in
k-classes. The proposed discrimination method proceeds by inference of
successive neighbouring. It is inductive, interactive, iterative and semi-directed. It
combines inductive techniques of discriminatory variables and neighbours search,
with the help of a similarity measure that takes into account the structure
(dependencies of variables) and the content (missing and unknown values) of
descriptions. It is based on the description space (inferior half-lattice) to guarantee
the coherency of descriptions and the computation of generalized descriptions.
Finally, the aim of our symbolic and numeric biological objects analysis is to
facilitate description and classification (for specialists) and identification (for
practitioners) of complex biological specimens on the Internet.
3.1
From data mining for classification and identification of taxa …
Statistical work in taxonomy has already been done for these processing tasks
since forty years with computer science tools for data analysis of phenetic attributevalue pairs (Sneath and Sokal 1973), i.e. flat attributes. All these numerical
approaches led to knowledge discovery from data mining in large databases
(Pankhurst 1991), (Fayyad et al. 1996). On the knowledge acquisition process,
artificial intelligence techniques (Frames) and object oriented languages (Smalltalk,
Java) brought also symbolic representational capacities for modelling more
complex specimen descriptions. At the same time, live memory capacities of
4
5
[http://www.gbif.org, visited on 05/10/2012]
[http://www.vbrant.eu, visited on 05/10/2012]
Noel Conruyt et al. / Knowledge Discovery for Biodiversity: from Data Mining to Sign Management
computers has grown fast to store more and more complex object descriptions. For
biological objects, Lebbe (1995) mentioned that there was a real need for
implementing representation models of the class/instance type, as well as following
knowledge acquisition methodologies, in order to process, refine and communicate
such evolving knowledge. As a result, we use Object-Attribute-Value
representations that give nowadays experts and knowledge engineers the
possibility to model morphological descriptions by structuring them in knowledge
bases rather than in databases. Knowledge bases are applications of Knowledge
Based Management Systems (KBMS) funded on a more qualitative reasoning than
databases: they focus on the quality of descriptions with a lot of dependant objects,
attributes and values rather than on the quantity of records (for statistical
significance) with independent variables. The difference comes also from the fact
that this type of knowledge bases enhances the individual level of description
(collection specimens) represented by individual cases rather than the conceptual
level of domain definition (taxa) represented in a record of a database or a line of a
data table.
The whole qualitative work that we have done in this context of specimen
description is to capture this morphological and individual complexity in an Iterative
Knowledge Base System (Conruyt and Grosser 2007). In fact, acquiring structured
specimen descriptions is particularly important in zoology because it gives sense to
the representation of entities in description trees with body subparts, viewpoints,
specializations of objects and values (figure 1).
Figure 1. A part of description tree of a specimen of coral (Stylophora subseriata)
This anthropomorphic and individual descriptive approach is less used in botanical
modelling of specimens because plants are mostly considered as species entities
representing populations with independent morphological characters (Operational
Taxonomic Units) (Dallwitz et al. 1995). This approach at the species level is
2
covered by another type of KBMS called Xper , which is a management system for
storage, editing, analysis and on-line distribution of descriptive data (Ung et al.
2
2010). A description made with Xper is represented by a set of character-states
describing one species (figure 2), whereas a description made with IKBS is
represented by a set of Object-Attribute-Value describing one specimen (figure 1).
These two knowledge acquisition processes are complementary: they have been
6
7
used on both corals and plants classification and identification tasks at University
6
[http://coraux.univ-reunion.fr/, visited on 05/10/2012]
Noel Conruyt et al. / Knowledge Discovery for Biodiversity: from Data Mining to Sign Management
2
of Reunion Island. Xper knowledge bases are built in intension (conceptual
description), starting from species description found in monographs and applying a
top-down approach for identification. IKBS knowledge bases are made in extension
(class description), for enhancing the collections of specimens found in museums
and applying a bottom-up approach for classification, then identification.
Figure 2. A description list of a species of plant (Trochetia granulata)
The first method is interesting when experts want to transmit an encyclopaedic
knowledge rather quickly and when taxa are well known and stable. Some
difficulties reveals once the domain knowledge evolves with new knowledge (a new
specimen description, a new character), and that consequently the application
needs to maintain its consistency (Pullan et al. 2000). The second method is more
time consuming for the expert who has to consider the intra-specific variation of
species and decide how many specimen descriptions he will store in the
knowledge base.
3.2
… to sign management for definition and description of specimens
For our coral species identification process, specimen entities to describe are
complex: they are altogether individuals, objects and populations. Colonies that are
found in museum collections are composed of corallites (objects) that form the
habitat of polyps (individuals). The colony (population of polyps) presents
polymorphic characters with conjunction of values (simultaneous presence of
states or values). It expresses intra-specific variability that depends on the
biological and geographical sampling context of colonies (reef zone, lightning,
profoundness, waves, etc.). Le Renard and Conruyt (1994) have specifically
designed descriptive logics for making descriptive models of these biological
entities. These descriptive logics of the Semiotic Web must not be confused with
the description logics of the Semantic Web. Description logics (OWL, RDF) are
the propositions of computer scientists that conceive syntactical formal
representations for expressing ontologies and descriptions on the business object
side of the machine’s interaction. Conversely, descriptive logics are propositions of
domain experts for guiding the building of ontologies that are meaningful and
understandable by the community. These other significant logics are user-centred;
they constitute the usage object side of the interaction with humans. First, it makes
sense that a description tree is better suited for visualization with dependent
7
[http://mahots.univ-reunion.fr/, visited on 05/10/2012]
Noel Conruyt et al. / Knowledge Discovery for Biodiversity: from Data Mining to Sign Management
objects and attributes (the absence of components causes the absence of subcomponents and the irrelevance of its attributes, see figure 1). Second, the
knowledge acquisition of these descriptions is made at a multi-level scale with
macroscopic and microscopic observations; these viewpoints are sources of error
(noisy characters, what to observe, influence of light under the binocular and the
microscope) and can lead to a lot of unknown and missing values. So, it is
important to establish visually the spatial level of observation by introducing these
viewpoints in the descriptive model (ecology, geography, morphology, molecular,
etc.). The structuration of the domain could also propose the temporal level of
observation of a specimen by introducing life cycles as another dimension of the
specimen description (i.e. larval, juvenile, adult). Third, modelling such ontologies
closes the world of possible descriptions. The expert defines what is observable in
the domain at the root of the description tree. All observed descriptions are then
comparable by using the same frame of reference. He can decide to update his
descriptive model by changing objects, attributes and values and then iterate on
the description process. Moreover, different persons can give their own description
of the same specimen in the case base. This was verified as a source of
robustness for identification purpose to avoid misinterpretations of end-users
(Conruyt and Grosser 2007). The design of a descriptive model with the help of
descriptive logics and the possibility to introduce inter-observer variation is what
makes sense of creating useful and usable knowledge bases, then used because
the service espouses the needs of end-users for teaching and learning
observational sciences. This communication process is called signification or
construction of the Sign. It is at the root of a new paradigm for utility, usability and
effective use of e-services in context called Sign management.
A Sign (figure 3) is the interpretation of an Object communicated by a Subject at a
given time and place, which takes into account its content (Data, fact, event), its
form (Information), and its meaning (Knowledge).
Figure 3. What is a Sign?
Sign management involves end-users with researchers and entrepreneurs for
making them participate to the design of the product/service that they want.
The problem that we have to face when making knowledge bases is that their
usefulness depends on the correct interpretation of questions that are proposed
by the system to obtain a good result of identification. Hence, in order to get a right
identification, it is necessary to acquire qualitative descriptions. But these
descriptions rely themselves on the observation guide that is proposed by the
descriptive model. Moreover, the definition of this ontology is dependent upon easy
visualization of descriptive logics. At last, the objects that are part of the
descriptive model must be explained in a thesaurus for them to be correctly
interpreted by targeted end-users. Behind each Object, there is a Subject that
models this Object and gives it an interpretation. In life sciences, these objects can
Noel Conruyt et al. / Knowledge Discovery for Biodiversity: from Data Mining to Sign Management
be shown to other interpreters and this communication between Subjects is
compulsory for sharing interpretations, and not only transmitting knowledge.
The challenge of Sign management for KDES is to involve all types of end-users in
the co-design of Sign bases for them to be really used (e-service). It is why we
emphasize the instantiation of a Living Lab in Teaching and Learning at University
of Reunion Island (Conruyt 2010) for sharing interpretations of objects and
specimens on the table rather than concepts and taxa in the head of subjects.
VALIDATION
Our knowledge management methodology has been applied in the domains of
Mascarene Islands corals classification and plants identification for several years
8
now. For our experts on corals of the Mascarene archipelago, the iterative process
of knowledge acquisition is the most important task for building a robust knowledge
base. IKBS brings them a reflexive tool for evaluating their observed descriptions in
regards to the evolution of their projections in observable descriptive models. For
families of coral knowledge bases, there are more than 30 iterations with IKBS
before the expert finds a robust ontology that well explains his interpretations of
observations. In Conruyt and Grosser (1999), we analysed the results of the
identification process with end-users. We have found that better identification
results are produced when we introduce other non-expert descriptions of the same
specimens in the initial case base. Consequently, the expert, aware of the
importance of transmitting his knowledge to other biologists, postulates more
precise and relevant characters that may be easier to observe and/or offer less
ambiguous values (easier to interpret) in his descriptive model. For example, he
will refine on the basis of mutually exclusive values, monosemic attributes, frames
of reference, warning signals, and enhanced illustrations on noisy questions.
Misinterpretations between end-users are also at the root of identification problems
9
for plant identification. As we can watch it in the film on the BACOMAR project on
an endemic family of the Mascarene Islands (Dombeyoideae), a lot of illustrations
are needed to explain the terminology and characters to observe when using
Xper2. This is the reason why it is mandatory to pass from knowledge transmission
with data mining to know-how sharing with sign management. A collaborative
workplace for teaching and learning will be the physical and virtual space for
exchanging interpretations of objects between the different actors that want to
manage biodiversity.
CONCLUSION
Since several years, we proposed a knowledge management method based on
top-down transmission of experts’ knowledge, i.e. acquisition of a descriptive model
and structured cases and then processing of these specimens’ descriptions with
decision trees and case-based reasoning. We designed tools to build knowledge
bases. But the fact is that Knowledge is transmitted with text, not shared with
multimedia, and there is a gap between interpretations of specialists and end-users
that prevents these lasts from getting the right identification. Today, we prefer to
deliver a sign management method for teaching and learning how to identify these
specimens on a Co-Design or Creativity Platform. This bottom-up approach is more
pragmatic and user-centred than the previous one because it implicates end-users
at will and is open to questions and answers. The role of experts is to show
amateurs how to observe, interpret and describe specimens in situ and on the
table. The responsibility of biomaticians (biologists and computer scientists) is to
store and share experts’ interpretations of their observation, i.e. know-how rather
than knowledge in sign bases with multimedia annotations for helping experts to
define terms, model their domain, and allow end-users to describe and identify
correctly the objects.
8
9
[http://coraux.univ-reunion.fr/spip.php?article101&lang=en, visited on 05/10/2012]
[ttp://www.canal-u.tv/video/universite_de_la_reunion_sun/bacomar_mahots.7598]
Noel Conruyt et al. / Knowledge Discovery for Biodiversity: from Data Mining to Sign Management
ACKNOWLEDGMENTS
We are grateful to the PO-FEDER 2-06 ICT measure for supporting our NEXTIC
and BACOMAR projects: [http://www.reunioneurope.org/UE_PO-FEDER.asp] and
to FP7 VIBRANT project [www.vbrant.eu] for giving us access to the International
Biodiversity Informatics community for enhancing new ideas and concepts.
REFERENCES
Breiman, L., Friedman, J. H., Olshen, R. A., and Stone, C. J., Classification and
Regression Trees. Belmont, Calif.: Wadsworth, 1984.
Cukier, K., Data, data everywhere, special report, The economist, 2010.
Conruyt, N., Grosser, D., Managing complex in knowledge environmental sciences,
Proceedings of the International Conference on Case-Based Reasoning
(ICCBR'99), 401–414, Munich, July 27-30, 1999.
Conruyt, N., Grosser, D., Knowledge management in environmental sciences with
IKBS: application to Systematics of Corals of the Mascarene Archipelago,
Selected Contributions in Data Analysis and Classification, Series: Studies in
Classification, Data Analysis, and Knowledge Organization, Brito, P., Bertrand,
P., Cucumel, G., De Carvalho, F. (Eds.), 333–344, Springer, 2007.
Conruyt, N., Sébastien, D., Payet, D., Geynet, Y., Caron, D., and Grosser D.,
Managing Insular Tropical Environment through Data and Knowledge Bases by
using Web Services: a case study on Corals and Herbarium of La Réunion
Island, 4th International Congress on Environmental Modelling and Software,
iEMSs’2008, M. Sànchez-Marrè, J. Béjar, J. Comas, A. Rizzoli and G. Guariso
(Eds.), volume 1, 264–271, Barcelona, 2008.
Conruyt, N., Sebastien, D., Vignes Lebbe, R., Cosadia, S., Touraivane, Moving
from Biodiversity Information Systems to Biodiversity Information Services, L.
Maurer, K. Tochtermann (Eds.): Information and Communication Technologies
for Biodiversity Conservation and Agriculture, Shaker Verlag, Aachen, 2010.
Conruyt, N., The Reunion Living Lab model: towards Sign Management on a
Creativity Platform, 1st European Living Labs Summer School, Collaborative
Innovation through Living Labs, August 25-27, Paris, France, 2010.
Dallwitz, M., Pain, T., Zurcher, E., User’s guide to the DELTA system, a general
system for coding taxonomic descriptions, CSIRO Div. Entomol., 1995.
Diday, E., New approaches in classification and data analysis, Studies in
classification, data analysis, and knowledge organization, NATO Asi Series.
Series F, Computer and Systems Sciences, Springer-Verlag, 1994.
e-Biosphere, book of abstracts, 2009. [http://www.e-biosphere09.org/]
Engeström, Y., Learning by expanding: an activity-theoretical approach to
developmental research, Orienta-Konsultit Oy, Helsinki, 1987.
Fayyad, U. M., Piatetsky-Shapiro, G., and Smyth, P., From Data Mining to
Knowledge Discovery: An Overview. In Advances in Knowledge Discovery and
Data Mining, eds. U. Fayyad, G. Piatetsky-Shapiro, P. Smyth, and R.
Uthurusamy, 1–30. Menlo Park, Calif.: AAAI Press, 1996.
Frey, J.G., The Semiotic Web: Don’t forget the user amongst all the semantics,
Proceedings of the First International Workshop on Understanding Web
Evolution, WebEvolve 2008, 27–34, Beijing, China, 2008.
Fukuda, K., Computer-enhanced knowledge discovery in environmental science,
Phd of the University of Canterbury, 2009.
Gibert, K., Spate, J., Sànchez-Marré, M., Athanasiadis, I., and Comas, J., Data
Mining for Environmental Systems. In Environmental Modelling, Software and
Decision Support: State of the art and new perspective, Jakeman, A., Voinov, A.,
Rizzoli, A. E., and Chen, S., (Eds.), 205–228, Elsevier, 2008.
Gibert, K., Sànchez-Marré, M., and Codina, V., Choosing the Right Data Mining
Technique: Classification of Methods and Intelligent Recommendation, 5th
International Congress on Environmental Modelling and Software, iEMSs’2010,
David A. Swayne, Wanhong Yang, A. A. Voinov, A. Rizzoli, T. Filatova (Eds.),
Ottawa, Canada, 2010. [http://www.iemss.org/iemss2010/proceedings.html]
Grosser, D., Conruyt, N., and Ralambondrainy, H., Identification with iterative
nearest neighbors using domain knowledge, Proceedings of the International
Noel Conruyt et al. / Knowledge Discovery for Biodiversity: from Data Mining to Sign Management
Congress on Tools for Identifying Biodiversity, Progress and problems,
BioIdentify, P. Luigi and R. Vignes Lebbe (eds.), Paris, September 20-22, 2010.
Lebbe J., Systématique et informatique. Systématique et biodiversité, Bourgoin T.
(Ed), Biosystema, 13:71–79, Paris, 1995.
Le Renard, J., and Conruyt, N., On the representation of observational data used
for classification and identification of natural objects, Studies in Classification,
Data Analysis and Knowledge Organization, New approaches in Classification
and Data Analysis, Diday, E., Lechevallier, Y., Shader, M., Bertrand, P.,
Burtschy, B. (Eds.), Springer Verlag, 308–315, Paris, 1994.
Mansur, A., Hossain, M. A., and Kuno, Y., Integration of Multiple Methods for Class
and Specific Object Recognition, chap 4, 841–849, Springer LNCS, Berlin, 2006.
Matile, L., Tassy, P., Goujet, D., Introduction à la systématique zoologique, In
Biosystema, vol. 1, Société Française de Systématique (Eds.), 1987.
Mayo, S. J., Allkin, R., Baker, W., Blagoderov, V., Brake, I., Clark, B., Govaerts, R.,
Godfray, C., Haigh, A., Hand, R., Harman, K., Jackson, M., Kilian, N., Kirkup, D.
W., Kitching, I., Knapp, S., Lewis, G. P., Malcolm, P., von Raab-Straube, E.,
Roberts, D. M., Scoble, M., Simpson, D. A., Smith, C., Smith, V., Villalba, S.,
Walley, L. and Wilkin, P., Alpha e-taxonomy: responses from the systematics
community to the biodiversity crisis, Kew Bulletin, 63:1–16, 2008.
Mitchell, T., Machine Learning, McGraw Hill, 1997.
Nature 455,1, 2008. [http://www.nature.com/news/specials/bigdata/index.html]
Pankhurst, R.J., Practical taxonomic computing. Cambridge University Press,
Cambridge, 1–202, 1991.
Peirce, C.S., Elements of Logic, In Collected Papers of C. S. Peirce (1839 - 1914),
C. H. Hartshone & P. Weiss, Eds. The Belknap Press, Harvard Univ. Press,
Cambridge, MA, 1965.
Piatetsky-Shapiro, G., Discovery, analysis, and presentation of strong rules,
Knowledge Discovery in Databases, AAAI/MIT Press, Cambridge, MA, 1991.
Poch, M., Comas, J., Rodríguez-Roda, I., Sànchez-Marrè, M., and Cortés, U.,
Designing and Building real Environmental Decision Support Systems,
Environmental Modelling & Software, 19(9): 857–873, 2004.
Pullan, M.R., Watson, M.F., Kennedy, J.B., Raguenaud, C., and Hyam, R., The
Prometheus Taxonomic Model: a practical approach to representing multiple
classifications, Taxon 49: 55–75, 2000.
Quinlan, J. R., C4.5: Programs for Machine Learning. Morgan Kaufmann Series in
Machine Learning, 1993.
Sarkar, I.N., Biodiversity Informatics: the emergence of a field, BMC Bioinformatics,
10(14), 2009.
Skyrme, D.J., Measuring The Value Of Knowledge, Business Intelligence, 1998.
Sneath, E., and Sokal, E., Numerical taxonomy, W.H. Freeman, San Francisco,
1973.
Uexküll, J. von, Theoretical Biology. (Transl. by D. L. MacKinnon. International
Library of Psychology, Philosophy and Scientific Method.) London: Kegan Paul,
Trench, Trubner & Co. xvi+362, 1926.
Ung, V., Dubus, G., Zaragueta-Bagils R., and Vignes-Lebbe, R., Xper2: introducing
e-taxonomy, Bioinformatics, 26 (5): 703-704, 2010.
[http://lis-upmc.snv.jussieu.fr/lis/?q=en/resources/software/xper2].
Winston, J. E., Describing Species: Practical Taxonomic Procedure for Biologists.
New York: Columbia University Press, 1999.
Zhu, X. and Davidson, I., Knowledge Discovery and Data Mining: Challenges and
Realities with Real World Data. Idea Group Inc, 2007.