Professional Documents
Culture Documents
© 2008 IEEE.
144
Authorized licensed use limited to: GOVERNMENT COLLEGE OF TECHNOLOGY. Downloaded on February 27, 2009 at 09:49 from IEEE Xplore. Restrictions apply.
2008 Second IEEE International Conference on Digital Ecosystems and Technologies (IEEE DEST 2008)
© 2008 IEEE.
* Interoperability support — in the simple case of consider- known as business agility. Business transformation is achieved
ing controlled vocabularies, there is enhanced interoperability sup- through efforts from the business and IT sides of the company.
port since different users/applications are using the same set of Thanks to transformation, operations can adapt to rapid strategic
terms. changes. When conditions change, goals and plans need to be ad-
* Support validation and verification testing of data (and apted, and the set of organizational assets — data, processes,
schemas) — if an ontology contains class descriptions, such as people — as well as everything that follows “change” need to be
“StanfordEmployee,” these definitions may be used as queries to modelled accordingly. Transformation can happen at three levels:
databases to discover what kind of coverage currently exists in 1. Business model transformation is a key strategic practice of
datasets. enabling and supporting — with suitable methods and techniques
* Encode entire test suites — an ontology may contain a num- — adaptation in an organizational environment through changes.
ber of definitions of terms, some instance definitions, and then in- 2. Data transformation is the process of redefining data based
clude a term definition that is considered to be a query: find all on some predefined rules that generally embed some kind of enter-
terms that meet the following conditions, for example. prise and business logic. The data values are redefined based on a
* Configuration support — class terms may be defined so that specific formula or technique.
they contain descriptions of what kinds of parts may be in a sys- 3. Process transformation is concerned with supporting the ad-
tem. aptation, evolution, and optimizations of organizational processes,
* Support structured, comparative, and customized search which are actually “ongoing” — think, for example, of “process
— for example, if one is looking for televisions, a class description improvement.”
for television may be obtained from an ontology, its properties Researchers who have studied business process modelling in
may be obtained (such as diagonal, price, manufacturer, etc.), and detail have identified, however, that most of the problems with
then a comparative presentation may be made of televisions by transformation, even the most practical ones, are of a conceptual
presenting the values of each of the properties. nature and so they could hardly be solved with technical innova-
* Exploit generalization/specialization information — if a tions, rather they need to be addressed at model and representation
search application finds that a user’s query generates too many an- level (6). According to the law of conservation of mass, (aka first
swers, one may dissect the query to see if any terms in it appear in law of thermodynamics), 'nothing is lost, and nothing is created,
an ontology, and, if so, then the search application may suggest everything is transformed'. (When applied to ‘data’, it has been
specializing that term. Digital ecosystems, tend to be chaotic in noted, this is however not entirely true, as data is created and can
nature. Among the overall benefits supplied by the development be lost) To be able to leverage the principles of transformation, and
and adoption of appropriate ontologies when designing open learn how to apply them in the context of social and information
world information environments, literature identifies: systems, the social sciences look at physics, aiming to capture
insights into the natural laws that may drive change. Information
* Improve the efficiency of reasoning plays a key role in the ability of an organism to adapt, respond and
* Consolidate and harmonize existing data/information survive to environmental 'changes'. Another important assertion of
* Provide an abstract, more simplified view of a system the key nature of transformation is found in the second law of ther-
* Create a consensual, unified view that can be used as a me- modynamics, which is essentially a general principle about the dy-
diation tool between different opinions namic of 'energy dispersion' (http://www.secondlaw.com) also
* Provide a formal specification known as the Law of Increased Entropy, which is a measure of un-
* Support integration of data, applications, and systems to usable energy. As usable energy decreases and unusable energy in-
help minimize design and planning errors caused by lack of do- creases, "entropy" increases. Entropy is also defined as a measure
main knowledge of randomness or chaos, as usable energy is irretrievably lost, dis-
organization, randomness and chaos increase. Entropy has been
Generally, it can be said that ontologies provide a consistent expressed and studied in different forms by different scientists,
view of reality, or of a model of reality, therefore they support bet- namely Canot, Clausius, Boltzmann, Gibbs . Each approach pro-
ter informed decisions and choices and allow collaborative intelli- poses a different interpretation relevant to its context: some define
gence to be built around them. entropy in terms of 'energy exchange', others in terms of 'ability to
do work', others in terms of 'measure of organization of a system'',
V. CAPTURING PATTERNS OF TRANSFORMATION: and so on, until more recent interpretations - entropy as semantic
THERMODYNAMICS APPROACH distance between two words, or as imprecision in the translation of
a term in two different languages. In the context of signal pro-
Anything that exists, that we know of, is constantly in flux, cessing and communication theory Shannon [7] famously related
transforming, moving, expanding, contracting, yielding and entropy to information and defined it as a "measure of the uncer-
morphing into the next thing When it comes to ‘digital ecosys- tainty associated with a random variable". Although some argued
tems’ such ongoing evolution is very evident and potentially dis- that the concept of 'entropy' applied to physical science as stated in
rupting to any business process which aims to rely on such envir- the second law is not directly comparable to the notion of entropy
onments. One of the main challenges of Business Intelligence is to in the communication theory context, the idea of information as
detect, and where possible to predict, patterns of change, to better 'available energy' is becoming increasingly acceptable in the in-
align internal organisational resources and processes that must formation age. David Wolpert, researcher with NASA has estab-
constantly be readjusted due to some change occurring in another lished a new line of research called 'Probability Collectives', which
part of the system, that inevitably reflects on all its components. determines that statistical physics and game theory 'are identical'
The ability of an organisation to transform, be flexible and adapt, [8].The Probability Collectives framework was initially designed
has always been viewed as a competitive advantage. Business to reduce flutter of an airplane wing , but its principles can be ap-
transformation is an executive management technique to align the plied in the social sciences by replacing the word "agent" with the
technology initiatives of a company more closely with its business word "player" throughout the maths, says the researcher. Wolpert
strategy and vision. The degree to which a company can imple- believes that the principles underlying information theory could be
ment new initiatives to support changes in business strategy is the binding glue among many disciplines "…Information theory
145
Authorized licensed use limited to: GOVERNMENT COLLEGE OF TECHNOLOGY. Downloaded on February 27, 2009 at 09:49 from IEEE Xplore. Restrictions apply.
2008 Second IEEE International Conference on Digital Ecosystems and Technologies (IEEE DEST 2008)
© 2008 IEEE.
has been extensively applied to many other disciplines, for ex- supported the principles advocated by Tom Stonier in the book "
ample, in the guise of the maximum entropy principle, it has found Information and the Internal Structure of the Universe" He said
great applicability in data processing and analysis. More generally, that information and organization are intimately related, that all or-
it is now recognized that there are a host of statistical inference ganized structures contain information, and no organized structure
techniques related to information theory which have proven very can exist without information content. The addition of information
powerful in many different fields” Wolpert says that in the future, to a system manifests itself by causing a system to become more
information theory's greatest new contributions will be to *relate* organized, or reorganized. If organization and disorganization are
disciplines and contribute to scientific convergence. Entropy is related to order and disorder, information has an inverse relation-
also used in mechanical statistics, quantum mechanics, economics, ship with entropy.” [14]. Richard Janow, a physicist, studies Shan-
and even in music. However it was not until Prigogine's times, (he non's entropy applied to the productivity of organizations , in par-
received the Nobel Prize for chemistry in 1977) who studied en- ticular with reference to their 'decision making abilities'. He intro-
tropy applied to chemistry to biology, and contributed to the emer- duces the concept of 'organizational entropy' and applies it to help
gence of chaos theory and complexity theory , that entropy has determine the degree of 'decision complexity', in the context of or-
been said to apply also to open systems, and not just to closed sys- ganisational growth, and to explain the increased time needed by
tems as previously believed. an organization to be capable of making decisions, in a non linear
proportion to its size. He writes: "In organizations, there are ana-
VI. SELF ORGANISATION logous flows of management- and application-related decisions,
and a modified Shannon model may be applicable. …Just as en-
The 'Law of maximum entropy production'(MEP), also known tropy for a communication channel measures the average informa-
as the 'law of spontaneous order' takes the second law a bit further. tion content (bits per symbol) of symbols transmitted over it or-
It says that 'entropy production is maximized at the fastest rate giv- ganizational entropy grows as a decision network grows even if
en the constraints' meaning that systems are inherently capable of the complexity of the tasks themselves remains unchanged. … Im-
selecting the most efficient route (which happens to coincide with portantly, the entropy grows fast enough to more than offset
the most organized system state) to compensate for disequilibrium growth in the total of individuals' capacities for making basic bin-
and reach stability (minimum entropy state).[9] ary decisions. … There appears to be a fundamental upper limit on
Maximum entropy is seen therefore as the main force behind the the total management decision rate that grows slower than linearly
'self –organisation' principle, which is the ability of a system to with the number of nodes, the maximum per capita management
find 'spontaneously' internal equilibrium and maximum stability decision rate therefore actually shrinks as the number of decision
[10]. Entropy based prediction and probability methods are ap- makers in the network grows". [15]. The findings above suggest
plied successfully in knowledge and data mining, where large data that when expanding (increasing in size) an organization is likely
sets and the number of unknown variable and missing data is sig- to become less efficient in terms of decision making capacity and
nificant. Tzannes and Noonan, of Tufts University, say that "The speed, given increased the 'dispersion rate' of its management , and
success of entropy methods is typically measured from an empiric- that measures to control 'structural entropy' should be taken.. Or-
al point of view, that is, they are applied to specific problems and ganizational entropy (disorganization, inefficiency, unusable en-
shown to be successful and often superior to classical methods. ergy) according to Janow grows with the size of the decision net-
This, of course, still leaves reasonable doubt in theoreticians work. 'The connection between entropy, information, and choice' is
minds as to the true merit of entropy methods and a fairly large long-standing, he says . "When the range of decision network
body of research has been dedicated to justifying their use from a states is large so is the entropy; the organization is then also
theoretical point of view . Typically: a relation between classical something of a general-purpose tool. Conversely, when entropy is
estimation theoretic methods and the entropy methods is shown small, the organization will probably do a prescribed set of spe-
that is meant to further validate the use of the entropy method. cialized tasks efficiently and others not at all. Large organizational
[12] entropy is the price of having the capabi1ity to execute a range of
complicated, multi-person decision/tasks; that ability impairs effi-
Entropy methods are sometimes used in combination with other ciency when doing simple tasks'"
methods, such as bayesian inference, helping the pre-processing
and data 'cleaning' required to support advanced business intelli- VII. SEMANTIC ENTROPY
gence algorithms “In data mining and data warehousing, data pre-
paration is often sized accurately at 60 to 90 percent of the effort - Entropy based algorithms are commonly adopted by scores of
and that is with structured data” says Lou Agosta, IBM business statistical and probabilistic methods developed in the attempt to
intelligence analyst, [12] "As business intelligence (BI) evolves establish and predict any given degree of certainty concerning the
from recounting the past to forecasting the future, unstructured in- dynamics of information in very large data sets, as well as on one
formation and enterprise search capabilities move to centre stage. of the most unpredictable, largely untested, fast growing informa-
...The emerging role of business intelligence systems is to alert de- tion environment, the semantic web. From the work of Philip Res-
cision-makers proactively about critical situations. This requires a nik [16] who in 1999 was already using statistical information to
number of search capabilities that are not usually associated with address the problems of ambiguity in natural language, to Dan
inferences and predictions based on the data in standard relational Melamed who uses semantic entropy can be applied to linguistic
databases, much less standard query and reporting tools that form translations, as a measure of semantic 'ambiguity and ‘uninformat-
the bulk of business intelligence applications. .. iveness’, entropy is used as measure of everything that you can't
Progress continues to be driven by advances in statistical and rule- quite put your fingers on.
based natural language processing (NLP), ontology (data model-
ling plus reasoning), information retrieval, machine learning, auto- VIII. ENTROPY AND ONTOLOGY
mated reasoning and knowledge sources (including lexicons and
frameworks for handling meaning).... . " GME is already adopted Today, knowledge management and decision support systems
by business intelligence solutions such as SAS [13]. Jim Nell, in enterprises are becoming increasingly 'ontology based', that is,
who worked on ISO standards for NIST Manufacturing Systems they rely on the existence of an accurate view of the world to mod-
146
Authorized licensed use limited to: GOVERNMENT COLLEGE OF TECHNOLOGY. Downloaded on February 27, 2009 at 09:49 from IEEE Xplore. Restrictions apply.
2008 Second IEEE International Conference on Digital Ecosystems and Technologies (IEEE DEST 2008)
© 2008 IEEE.
el the relevant supporting systems. Specialised information and de- interfaces capable of simulating ‘change’ to validate our hypothes-
cision support systems depend on correspondingly specialised on- is.
tologies: medicine, aerospace, civil engineering. Pefferly, Jaeger
and Lo [17] in a recent study make the case for taking into account REFERENCES
'entropy' as well as other variables, when building ontology based
knowledge management systems They say "when describing in- [1] Vakas Duong Grefenstette The emulation of social institutions as a
formation based systems, statistical measures are a necessity; yet method of coevolution Genetic And Evolutionary Computation
Conference archive Proceedings of the 2005 conference on Genetic
very few ontology based standards mention quantifiable measures
and evolutionary computation
such as entropy, data encapsulation, complexity, efficiency, evolu- [2] Jacob, Becker. Shapira1 Levine Bacterial linguistic communica-
tion, or redundancy". tion and social intelligence, TRENDS in Microbiology Vol.12 No.8
August 2004 star.tau.ac.il/~eshel/papers/Trends-published.pdf
[3] Almulla M., Szuba T.: Universal Formal Model of Collective Intel-
ligence and Its IQ Measure. Springer Verlaghttp://www.springerlink.-
IX. VALIDATION OF THE APPROACH com/content/jvk8rwx6u1xxd39l/
[4] Holzer Distributed Operations Transformation Trends. Newsletter of
There is increasing evidence that 'entropy' measurements can the Office of Force. Transformation. U.S. Department of Defense,
be applied successfully in calculating and predicting a range of February 2004.
variables. In the Gene Ontology project (GO), an effort to create a [5] McGuinness, “Ontologies Come of Age.” In Spinning the Semantic
controlled terminology for labelling gene functions in a more pre- Web: Bringing the World Wide Web to Its Full Potential. MIT Press,
cise, reliable, computer-readable manner, researchers are studying 2003.
[6] Murzek Marion, and Gerhard Kramler. “Business Process Model
methods to produce annotations of gene function currently per-
Transformation Issues: The Top 7 Adversaries Encountered at Defin-
formed by highly trained biologists who read the literature and se- ing Model Transformations,” 2007.
lect appropriate codes with statistical natural language processing [7] Shannon, C ‘Mathematical Theory of Communication’
techniques. They compare three document classification methods http://plan9.bell-labs.com/cm/ms/what/shannonday/shannon1948.pdf
(maximum entropy, naïve Bayes classification, and nearest-neigh- [8] D. Wolpert, Probability Collectives Nasa
bor classification) and conclude that maximum entropy modeling http://ase.arc.nasa.gov/projects/probcol/publications.php
[9] Minimum entropy principle http://www.entropylaw.com/index.html
outperforms the other methods and achieves an accuracy of 72%
[10] Max Entropy State
when ascertaining the function discussed within an abstract. The http://www.entropylaw.com/thermoevolution10.html
maximum entropy method, the researchers say, provides confid- [11] Tsannes, Noonan,. On a Relation Between the Principle of Minim-
ence measures that correlate well with performance. [18] Pefferly um Relative Entropy and Maximum Likelihood Estimation IEEE In-
and Jaeger proposed metrics has recently been applied successfully ternational Symposium on Circuits and Systems, 1990....http://ieeex-
to measuring both complexity and information, differential met- plore.ieee.org/iel5/143/3356/00112235.pdf
rics for measuring productivity, fraud detection (in credit cards, in- [12] Lou Agosta IBM Business Intelligence Searches for Emerging Trends,
surance, business transactions), ontology complexity as well as DB DRM Review article
http://www.dmreview.com/editorial/dmreview/print_action.cfm?art-
construction, building cellular towers and/or network connection icleId=1052933 as retrieved on 2 Sep 2007 13:26:29 GMT.
hot-spots, evaluating trends in economic situations,evaluating the [13] SAS Documentation
worth of an IT system, spotting fraudulent trends in networks, al- http://support.sas.com/documentation/whatsnew/91x/etsugwhatsnew
locating financial resources - stocks as well as operational commit- 900.htm
ments. Some results of the above application of the metrics are in [14] J.Nell, EI3-IC Knowledge RepresentationWorkshop Proceedings
publication [19] www.cimosa.de/EI3-IC/WS1_abs.htm
[15] R. Janow, 'Shannon entropy applied to productivity of organiza-
tions , IEEE 1988
ieeeexplore.ieee.org/iel5/8871/28027/01252225.pdf
X. CONCLUSION AND FUTURE WORK [16] P.Resnik. 1999. Semantic Similarity in a Taxonomy: An Informa-
tion-Based Measure and its Application to Problems of Ambiguity in
Although human understanding of how the second law of ther- Natural Language.Journal Artificial Intelligence Research.
[17] Pefferly Jaeger and Lo Metrics for Objective Ontology Evaluations
modynamics applies to digital ecosystems is still limited, having
http://www.springerlink.com/content/7473026p43x03t2g/
researched and demonstrated through literature review the relev-
ance of the second law of thermodynamics to Business Intelli- [18] Raychaudhuri, Chang, Sutphin, Altman Associating Genes with
gence, we put forward the hypothesis that 'entropy' as a measure of Gene Ontology Codes Using a Maximum Entropy Analysis of Bio-
change and transformation is a factor in models of reality, and as medical Literature Departments of 1Genetics and 2Radiation Onco-
such, it should be represented accordingly, following the logical logy, Stanford University,
[19] R.Pefferly, A Shannon Econometric Approach to Measuring Fre-
construct below:
quency and Severity Phenomena (forthcoming )
IF
ontology represents reality (or a bounded subset thereof),
and reality changes/transforms constantly,
and entropy provides a measure to calculate/predict such changes
THEN
ontology engineering should use entropy measurement (as well as possibly
other methods to calculate change and transformation of the realities it rep-
resent)
147
Authorized licensed use limited to: GOVERNMENT COLLEGE OF TECHNOLOGY. Downloaded on February 27, 2009 at 09:49 from IEEE Xplore. Restrictions apply.