SEMANTIC APPLICATION DESIGN
Philippe Larvet
Alcatel-Lucent Bell Labs, Centre de Villarceaux, 91620 Nozay, France
Keywords: Application design, software component, semantic component.
Abstract: This paper presents a process to determine the design of an application by building and optimizing the
network of semantic software components that compose the application. An application has to implement a
given specification. We consider this specification is made of atomic requirements, logically linked
together. Each requirement is expressed in natural language: this expression is seen as the semantic
description of the requirement. Off-the-shelf components from which we want to build the application can
also be described through a semantic description. We consider a component implements a requirement if the
"semantic distance" between their two semantic descriptions is minimal. Consequently, designing an
application consists of building and optimizing the logical network of all semantic optimal couples
"requirement-component". The paper presents such a building and optimization automatic process, whose
development and improvement are still in progress, and whose main advantage is to systematically derive
the discovery and assembly of software components from the written specification of the application.
1 PROBLEM OF APPLICATION
DESIGN
Software application design is traditionally a
complex activity. According to the accepted
definitions used as references in the scope of object-
oriented and component-based application
development, and according to Grady Booch
(Booch, 2007), design is "that stage of a system that
describes how the system will be implemented, at a
logical level above actual code. For design,
strategic and tactical decisions are made to meet the
required functional and quality requirements of the
system. The results of this stage are represented by
design-level models: static view, state machine view,
and interaction view." The activity of design leads to
the architecture of the application, which is "the
organizational structure of a system, including its
decomposition into components, their connectivity,
interaction mechanisms, and the guiding principles
that inform the design of the system." (Rumbaugh,
Booch, Jacobson, 1999).
Many authors have described several methods
to guide the building of component-based
applications (Bordeleau, 2005; Chusho, 2000;
Kirtland, 1998) but except within the field of
semantic web services (Narayanan, 2002), it seems
an automatic semantic-oriented process has not been
considered as a serious approach to design.
Traditional component-based development
approaches have two main drawbacks: they are often
fully manual, and the process of finding and
assembling the right components is not directly
derived from the text of the written specification.
Notice that research in semantic web services
(Narayanan, 2002; Patel-Schneider, 2002) proposes
a semantic-oriented design approach, by using a
logic-based approach to the specification of
semantics, whereas the approach presented in this
paper uses natural language as the basis for
specifying semantics.
An application has to rely on a given
specification. We consider this specification exists
under the form of a natural language, informal text,
that describes functional and non functional
requirements the application has to cover. We have
made three main assumptions in this paper:
- a software application can be built by
assembling off-the-shelf components;
- the determination of components can be derived
from the semantic analysis of the requirements;
- application design, i.e. architecture of solution,
can be derived from the architecture of the
problem, i.e. from relationships between
requirements.
47
Larvet P. (2008).
SEMANTIC APPLICATION DESIGN.
In Proceedings of the Third International Conference on Evaluation of Novel Approaches to Software Engineering, pages 47-55
DOI: 10.5220/0001761900470055
Copyright
c
SciTePress
2 THE PROPOSED PROCESS
We see an application as a set of inter-related
components. Each component has a functionality,
expressed as a set of functions, and encapsulates and
manages its own data: this is the component
paradigm, derived from object-orientation and
today's standard of development.
Let us consider we have at our disposal many
small off-the-shelf components, stored in
appropriate component repositories. Each one covers
a precise elementary function, an atom of
functionality – for example file management,
database access, GUI display mechanisms, text
translation, HTML pages reading from URLs,
elementary functions for text processing, etc. More
simply than the concept of semantic component
(Kaiya, 2005; Sjachyn, 2006; Hai, 2006), we
propose each component is described through a
semantic card which contains notably the goal of the
component, expressed in natural language form and
describing clearly what the component really does,
what its functions are and which data it manipulates.
Through an appropriate process we expose in
detail below, the meaning of the sentence
representing the component's goal - its semantics –
can be determine and expressed in terms of an
appropriate computable data structure. Thus, the
idea is to mark every semantic atom of functionality
with their appropriate semantic data structure.
We also have at our disposal a specification
document containing requirements describing what
the application will do, what its functional and non-
functional features are. The requirements are a set of
sentences expressed in natural language. Each
sentence has a meaning which can be found out by
using the same process. Each sentence, i.e. each
piece of specification, each atom of requirement, can
therefore be evaluated and marked, and each
sentence will receive its own semantic data.
Notice that this process is different than an
ontology-based requirement analysis approach
(Kaiya, 2005) – an ontology (McGuinness, 2004) is
a formal description of the concepts manipulated in
a given domain and of relationships between these
concepts. Here, no external ontology is used to help
requirements analysis, because semantics is
extracted from the text itself.
Sentences that compose requirements are
logically linked to each other. Then, it is possible to
determine a requirement network by scanning links
between requirement atoms: this browsing will
determine the structure of the 'specification
molecule' – the molecule that describes the problem.
Analyzing lots of specifications within the
context of numerous industrial projects developed
with an object-oriented approach (Larvet, 1994) has
led us to observe that a link between two different
requirements in the specification always leads to a
link between the classes implementing these
requirements. Indeed, two pieces of requirement are
linked to each other when they both talk about a
given data, constraint, functionality or feature of the
targeted application. Then, the same link exists
between the components implementing these
requirements.
Consequently, it makes sense to consider that
links between the bricks of the problem have a
similar correspondence to links between the blocks
of the solution. In other terms, problem structure –
'specification molecule' – is isomorphic to solution
structure – 'design molecule'.
Our proposed process consists of three steps:
1. finding the components whose semantic
distance is the shortest with semantic atoms of
requirements;
2. organizing these components in order to
constitute the 'solution molecule', i.e. the initial
architecture of the application – this initial
design being made by replicating the problem
molecule and using solution atoms instead of
problem atoms – but these kinds of atoms do
not have exactly the same nature, so the initial
component interaction model has to be
optimized; and
3. optimizing the structure of solution molecule in
order to determine the best component
interaction model.
Within this approach, the initial component
interaction model – corresponding to the initial
design of the future application - is built from
relationships between application's requirements: an
association between two requirements will
determine an association between the two
components that cover these requirements.
3 SEMANTIC CARDS FOR
COMPONENTS
Semantic cards (semCards) formally describe the
small off-the-shelf components that are used to build
applications. Each semCard contains the goal of the
component and the list of its public functions with
their input and output data. We propose a semCard
has an XML representation where input and output
data are described with three main attributes:
ENASE 2008 - International Conference on Evaluation of Novel Approaches to Software Engineering
48
1. a data name;
2. a concept associated with the data, expressed
in reference to a word defined in an external
dictionary or thesaurus, in order to specify
the semantics of the data; here, the concept
belongs to a domain addressed by the
component and whose name is mentioned in
semCard's header; and
3. a semantic tag, or semTag, of the data, which
represents a stereotype of a semantic data
type and specifies the nature of the data
(Larvet, 2006); this semTag will be useful to
determine and optimize components'
interactions.
The semantics of the operations' goals is defined
with precise rules that help to write terse and non
ambiguous expressions:
- goals are expressed in natural language, using
specific words;
- these words belong to 'lists of concepts' that are
embedded in the semCard and summarize the
pertinent words to be used to write goals; and
- words composing 'lists of concepts' are defined
in external dictionaries and belong to related
domains that are referenced in the semCard.
Ontologies could be used to summarize and
formalize the definitions of concepts and domains,
but this is not mandatory. RDF (Patel-Schneider &
Siméon, 2002) or OWL (McGuinness, 2004) are
convenient to depict such ontologies, because they
are standard and well-tooled languages, but simple
ad-hoc appropriate XML files containing word
definitions and domain descriptions are also suitable.
Here is, as an example, the semCard for an RSS-
feed-accessor component:
<semCard>
<URL>http://xxx.xx.xxx.x/components/RSS
/RSS_Component.asmx</URL>
<component name="RSS">
<domains>
<domain name="RSS">
<concepts list="RSS, RSS_feed,
URL, news" /></domain>
<domain name="News">
<concepts list="news, title,
titles, description, article, text,
News Agency" /></domain>
</domains>
<operation name="getAllTitles">
<goal>The goal of the operation
getAllTitles is to deliver the titles
of all the news of the RSS feed
addressed by a given URL.</goal>
<input name="URL_RSS"
concept="RSS#URL" semTag="URL" />
<output name="titles"
concept="News#title" semTag="text" />
</operation>
<operation
name="getDescriptionOfTitle">
<goal>The goal of the operation
getDescriptionOfTitle is to deliver the
description of the given title of one
news of the RSS feed addressed by a
given URL.</goal>
<input name="URL_RSS"
concept="RSS#URL" semTag="URL" />
<input name="title"
concept="News#title"
semTag="short_text" />
<output name="description_
of_title" concept="News#description"
semTag="text" />
</operation>
</component>
</semCard>
4 DETERMINING THE
MEANING OF SENTENCES
One key to our process is the possibility to compare
the meaning of a requirement extracted from the
specification document with a component's goal,
written in the component's semCard. This
comparison is done in order to be able to choose this
component because it is intended to cover the
requirement.
The key to this comparison is the ability to
determine the meaning of a text. We consider this
meaning is made up of the concatenation of
elementary meanings of all the pertinent terms that
compose the text. The ability to compare the
meaning of two different texts implies the ability to
compare two different terms and to determine
whether they are semantically close or not.
Important works have been done related to
semantic proximity in natural language expressions
(Khaitan, 2006; Corley, 2005; Guha et al. 2003;
Mayfield and Finin, 2003; Guarino et al., 1999;
Evans and Zhai, 1996). The novelty of our approach
is to propose a way to express the meaning of an
elementary term in order to process a comparison
with another term. To do so, we build a "synonym
vector" with the synonyms of the term that can be
found in a thesaurus, and we call it a synVector.
For example, the synVector of "battle" is:
battle = {fight, clash, combat,
encounter, skirmish, scuffle, mêlée,
conflict, confrontation, fracas, fray,
SEMANTIC APPLICATION DESIGN
49
action; struggle, crusade, war,
campaign, drive, wrangle, engagement}
Other examples:
war = {conflict, combat, warfare,
fighting, confrontation, hostilities,
battle; campaign, struggle, crusade;
competition, rivalry, feud}
peace = {concord, peacetime,
amity, harmony, armistice,
reconciliation, ceasefire, accord,
goodwill; agreement, pact,
pacification, neutrality, negotiation}
Table 1: Functions defined on synVectors.
Notation Semantics
synV(word) synVector for the term 'word'
card(V1) cardinal(vector V1)
common(V1, V2)
{syn
1
, syn
2
, …, syn
n
} | syn
i
V1 and
syn
i
V2
Example:
card(common(synV("battle"),
synV("war"))) = 9
avg(V1, V2) average(cardinals(V1,V2))
semProx(T1, T2)
100 * card(common(synV(T1),
synV(T2))) / avg(synV(T1),
synV(T2))
phraseVector(sen
tence)
{synV(Ti)} | Ti {T1, T2, … Tn} and
Ti = pertinent word of <sentence>
semVector(sente
nce1, sentence2)
{best values of comparisons among
all couples synV(T1),synV(T2) | T1
sentence1 and T2 sentence2}
diff(semV(req1,re
q2),
semV(req1,req3))
abs( semVector(req1,req2)
semVector(req1, req3) )
The concept of semantic proximity between two
terms T1 and T2 = semProx(T1,T2), gives a ratio
taking into account common synonyms within the
two synVectors of T1 and T2. If semProx is greater
than a given value A (for instance 50) or close to
100, we consider the two terms are semantically
close. For example,
semProx("battle", "war")
= 100 * 9 / 0.5 * (19 + 13) = 56.25. In
other words, in the union of synonyms sets for
"battle" and "war", 56% of the elements are found in
duplicate. Inversely, if semProx is less than a given
value B (for instance 10) or close to zero, the two
terms are semantically distant. Obviously, for
instance,
semProx("war", "peace") = 0.
Values of levels A and B can be "tuned",
according to the category of texts to be processed.
The determination of the meaning of a given
sentence is made as follows:
- the sentence is analyzed and pertinent words are
extracted – non-pertinent words like articles,
prepositions, conjunctions, etc, are ignored;
- for each pertinent word, a synVector is built;
- a vector of vectors for the whole sentence - a
phraseVector - is built by assembling all
synVectors of pertinent words contained in the
sentence, as shown in Figure 1.
Figure 1: Building a semVector from phraseVectors of
two sentences.
For example, let us build a phraseVector for the
following requirement, extracted from the
specification of a Call Management System:
requirement = The caller makes a
call to a receiver by creating a
message that contains the call subject,
submitted to the receiver at the same
time
Pertinent terms are: caller, call, make a call,
receiver, message, subject, submit.
The phraseVector for this requirement is the
concatenation of the following synVectors:
synV(caller) = {phone caller,
telephone caller, visitor, guest,
company}(5)
synV(call) = {phone call, telephone
call, buzz, bell, ring; demand,
request, plea, appeal, bid,
invitation}(11)
synV(make a call) = {phone, make a
demand, send a request}(3)
synV(receiver) = {recipient, heir,
addressee, beneficiary, inheritor,
heritor}(6)
synV(message) = {communication,
memo, memorandum, note, letter,
missive, dispatch}(7)
ENASE 2008 - International Conference on Evaluation of Novel Approaches to Software Engineering
50
synV(subject) = {topic, theme,
focus, subject matter, area under
discussion, question, issue, matter,
business, substance, text; field,
study, discipline, area}(15)
synV(submit) = {offer, present,
propose, suggest, tender}(5)
phraseVector(requirement) =
{synV(caller), synV(call), synV(make a
call), synV(receiver), synV(message),
synV(subject), synV(submit))
The comparison of three sentences S1, S2 and
S3 is made by comparing their phraseVectors (see
Figure 1). This comparison builds a result that will
be used to calculate the semantic distance between
the sentences. Let us detail the phraseVectors
comparison steps:
- internal synVectors of the two phraseVectors
are compared two by two – this means every
synVector in S1 is compared to every one in S2
and S3;
- a semantic proximity (semProx) is calculated
for each pair;
- the best values of semProx among all
comparisons are kept in an ordered external
semantic vector, a semVector, as a result of the
comparison; then
- the comparison of semVectors for sentences S1
and S2, and for S1and S3, allows to determine
whether S1 is semantically closer to S2 or S3.
The search of components that cover a given
specification follows the phraseVector approach:
- phraseVectors of requirements are built;
- phraseVectors of components' goals are built;
- phraseVectors are compared and the
corresponding semVectors are built for every
pair requirement-component; and
- the best semVectors are kept and help to
determine the components that are able to fulfill
the requirements.
5 DETERMINING THE
PROBLEM NETWORK
The requirement network summarizes and represents
the links between the requirements.
The phraseVector approach reveals the links
between the requirement atoms and helps the
building of 'problem molecule': the sentences of the
specification are semantically compared two by two,
phraseVectors are built and semVectors are
calculated.
Figure 2: Building the requirement network.
The result, for each requirement, is a set of
vectors that represent the links, in terms of semantic
distance, of each requirement with respect to the
others. We can "tune" the level of this semantic
distance to keep only the "best" semVectors in terms
of semantic proximity, i.e. the most semantically
pertinent links for a given requirement. This means
each requirement has a limited number of
semantically closest other requirements; in other
terms, a requirement can be formally described by a
limited set of semVectors that represent the
semantically closest other requirements.
The links can be represented in a 2D or 3D
space; the aim is to get a convenient model of the
problem, i.e. a representation we can communicate
and we can structurally compare to another. On this
model, we make graphically appear the links
between requirements. Only the best links are kept,
i.e. the links whose semVector value is larger.
For example, Req2 is linked with Req3 and
Req5, but
semVector(Req2,Req5)>semVector(Req2,Req3)
this means the semVector resulting of the
comparison between Req2 and Req5 is greater than
the semVector resulting of the comparison between
Req2 and Req3, then only the link Req2-Req5 will
be kept on the final model. This is a question of
optimization. Tuning the model is possible by
determining the maximum acceptable gap between
two semVectors.
6 BUILDING A PRIMARY
SOLUTION NETWORK
We assume that the structure of the solution, i.e. the
architecture of the design, is isomorphic to the
SEMANTIC APPLICATION DESIGN
51
structure of the problem. Solution molecule has the
same spatial structure as problem molecule, although
they do not contain and use the same kinds of atoms:
problem atoms are requirements, solution atoms are
components. Problem atoms are linked together
because they share the same concepts and address
the same requirements, solution atoms are linked
together because they share or exchange the same
data, however the network that links the
requirements together contains the same paths as the
network of the solution.
The problem now consists of finding the
components whose semantic distance is the shortest
from the semantic atoms of requirements, and
organizing these components in order to constitute
the solution molecule, i.e. the architecture of the
application that will suitably solve the problem
expressed in the specification document.
To build this organization, we will apply the
following steps:
- find the components that cover the requirements
by using the semVector approach; this will build
a list of components, not yet linked together (see
Figure 3);
- replicate the structure of the problem molecule
inside the components list using solution atoms
instead of problem atoms, i.e. by attaching to the
corresponding components the links between the
requirements they fulfill; this will build a rough
version of the solution molecule;
- and finally optimize this primary version in order
to determine the best structure for the solution
molecule. This will become the final architecture
for the application.
Figure 3: SemVectors help to determine which
components fulfill which requirements.
The optimization process will use semantic tags
attached to data descriptions of components'
operations to determine and optimize interactions
between components. The final result of this process
is an interaction diagram showing coupling and
interdependencies between components.
Replicating requirements links inside
components' structure associates components in the
same way requirements are associated in the
specification; but of course these associations are not
all valid: the fact that two requirements share the
same concepts does not necessarily imply the two
corresponding components have an interaction.
The role of the optimization process is to keep
only the most useful of the links inherited from the
problem molecule, i.e. the associations
corresponding to actual data exchanges between
components.
7 OPTIMIZING THE SOLUTION
NETWORK
In order to automatically determine real connections
corresponding to actual data exchanges between
components, we use the semantic tags (semTags)
added as semantic metadata to inputs and output of
components' operations (Larvet, 2006).
If these semTags are suitably chosen and set,
components can be connected and their connectivity
can be formally expressed.
For example, if output of Comp1.operationA
semantically fits with input of Comp2.operationB,
then Comp1 can be connected to Comp2 through the
link "output of A" to "input of B".
So, we can write:
out_A=Comp1.operationA(parameters);
out_B=Comp2.operationB(out_A);
or, more directly:
out_B=Comp2.operationB(Comp1.operationA
(parameters));
This means the two connected data have the
same semantic "dimension", i.e. they are process-
compatible; they share not only the same data type,
but the same nature of data. SemTags express
semantic data types and are similar to UML tagged
values (Rumbaugh et al., 1999); they are attached to
inputs and outputs within the semCards, and ensure
the consistency of components' interfaces; for this
reason they are important elements for optimizing
components interactions (see for instance semtags in
RSS-feed-accessor semCard, in paragraph 3.)
7.1 Automating the Optimization of
Solution Network
An example will help us to describe the process that
takes into account semantic tags in order to build an
ENASE 2008 - International Conference on Evaluation of Novel Approaches to Software Engineering
52
automatic assembly of components. Suppose we
want to produce a translated version of a news feed.
This requirement is expressed in natural language in
the specification, and the semVector-plus-
component-discovery approach has allocated two
components to this requirement: a RSS-accessor
component and a Translator component.
The RSS component aims at gathering information
from RSS feeds accessible via Internet, and its
interface contains two operations:
getAllTitles()
gets all the main titles of the feed for a given URL, and
getDescriptionOfTitle() gets the text of the
short article for this title.
The Translator component is a classical one whose
operation
translate() transforms a text (given as
an input parameter) written in a given source language
(input parameter) into a translated text (output) written
in a destination language (input parameter).
The problem is to assemble automatically and
logically these two components, i.e. their three
operations (see Figure 4) in order to fulfill the
original requirement: provide a translated version of
a news feed.
Figure 4: How to assemble these 3 operations?
The first key is to consider semantic tags as inputs
and outputs of operations, instead of data. Then, some
possible connectivities appear (see Figure 5), but not
precisely enough to make a fully consistent
composition.
Figure 5: Possible connections (in blue) appear by
considering semantic tags instead of data names.
The second key is to consider the main output of
the targeted component assembly in order to find
which operations can provide its inputs, and to
iterate the process for these operations: search which
other operations can provide their inputs. Then, we
go back progressively from the main output to the
input data necessary to produce it, and in doing this,
we automatically assemble the different operations
by linking their outputs and inputs.
At the same time, the links are stored in a FILO
(first in, last out) stack under the form of pseudo-
code expressing the operation calls. At the end of
this process, the content of the stack represents the
correct interactions between the components.
The main output of the component assembly is
given by the expression of the original requirement.
For our example, a translated version is wished: the
main output is a translated text, i.e. the output of the
operation
Translator.translate(). We can push
this main output in the stack, expressed as the
"return" of the function represented by the targeted
component assembly:
translated_text =
Translator.translate(text_to_trans
late, src_lang, dest_lang);
return translated_text;
Let us go back now to the inputs of this
operation, whose semantic tags are "language",
"language" and "text". A data with a semantic tag
"text" is provided by the operation
RSS.getDescriptionOfTitle().
Then, we can connect this operation to
Translator.translate().We can add the call
to the operation
RSS.getDescriptionOf-
Title() in the stack, linking with
Translator.translate() through the name of
the exchanged parameter:
text_to_translate =
RSS.getDescriptionOfTitle(site_add
ress, title);
translated_text =
Translator.translate(text_to_trans
late, src_lang, dest_lang);
return translated_text;
Now, let us go back to the inputs of
RSS.getDescriptionOfTitle(), whose semantic
tags are "URL" and "title". A data with a semantic tag
"title" is provided by the operation
RSS.getAllTitles().
So, we can also connect these two operations by
pushing a new operation call in the stack:
titles =
RSS.getRSSTitles(adr_site);
text_to_translate =
RSS.getDescriptionOfTitle(site_add
ress, title);
SEMANTIC APPLICATION DESIGN
53
translated_text =
Translator.translate(text_to_trans
late, src_lang, dest_lang);
return translated_text;
With all the components allocated to the
original requirement being used and connected
together, the stack now contains the general texture of
the component assembly, under the form of a nearly
executable pseudo-code. However, this pseudo-code
must be refined before it can be executed:
- the data types must be taken into account; for
example,
RSS.getAllTitles() returns an
array of Strings and not a single String;
- the names of some parameters can be solved
through their semantics, i.e. with the help of their
semTags: for instance, "adr_site" and
"site_address" recover the same concept and
have the same semTag;
- some other parameters can be solved with some
useful information contained in the original
requirement; for example, if the requirement
specifies a french translation, then the parameter
"dest_lang" of the operation
Translator.translate() has to be set to
"french"; and
- some additional components or operations can
be used to solve other parameters; for example,
the parameter "src_lang" can be set by using a
utility component, a "Language Finder", to
automatically determine the source language of a
given text, or an operation
getSourceLanguage() on the RSS feed
component.
A specific module, whose detailed description is
outside the scope of this paper, makes these
refinements in order to complete the pseudo-code:
Vector ComponentAssembly(String
site_address) {
Vector result;
titles =
RSS.getAllTitles(site_address);
foreach title in titles {
text_to_translate =
RSS.getDescriptionOfTitle(site_add
ress, title);
source_lang =
LanguageFinder.getLanguage(text_to
_translate);
translated_text =
Translator.Translate(text_to_tra
nslate, source_lang, "french");
result.add(title + translated_
text);
}
return result;
}
This pseudo-code can finally be transformed into
an executable Java file for example, in order to test
the validity of the component assembly produced by
the optimization process.
The final interaction diagram between the
components, obtained as a result of the optimization
process, can be considered as a first draft of the
design of the future application. The interest of this
draft is to be delivered with a quasi-executable
pseudo-code allowing validation tests of the
architecture of the future application.
8 CONCLUSIONS
The paper has described an application of a natural
language (NL) technology combined with a
component-composition optimization process in
order to allow the automatic construction of software
applications. We have presented an original but
partly operational process to determine the meaning
of a NL text, and to use this meaning to find the
right components fulfilling original NL-expressed
requirements of an application specification. This
process leads to an initial architectural structure of
the targeted application, optimizable with a
complementary process in order to get an acceptable
and testable draft of the application design.
Among some advantages of this approach, notice
that it is performed rapidly and fully automatically,
it works directly from the original application
requirements and delivers a quasi-executable
pseudo-code as a useful sub-product allowing a
validation of the future application's architecture.
Moreover, traceability between requirements and
architecture is guaranteed.
Still in progress, the process has to be improved
and refined. An important part of the future work is
to do more complete and rigorous experimentation,
validation, and perhaps tuning.
REFERENCES
Booch G., 2007. "Object-Oriented Analysis and Design
with Applications", 3rd Edition – Cased, Addison-
Wesley (2007), ISBN 9780201895513
Bordeleau F., Hermeling M., 2005. "Model-Driven
Development for Component-Based Application
Portability", COTS Journal, August 2005
Chusho T., Ishigure I., Konda N., Iwata T., 2000.
"Component-based application development on
architecture of a model, UI and components," apsec, p.
ENASE 2008 - International Conference on Evaluation of Novel Approaches to Software Engineering
54
349, Seventh Asia-Pacific Software Engineering
Conference (APSEC'00).
Corley C. and Mihalcea R., 2005. "Measuring the
Semantic Similarity of Texts". Proceedings of the
ACL Workshop on Empirical Modeling of Semantic
Equivalence and Entailment, page 1318, Ann Arbor.
Evans D. and Zhai C., 1996. "Nounphrase analysis in
unrestricted text for information Retrieval",
Proceedings of the 34th Annual Meeting of the
Association for Computational Linguistics, 1996.
Guarino N., Masolo C., Vetere G., 1999. "OntoSeek:
Content-Based Access to the Web", IEEE Intelligent
Systems, Vol. 14, No. 3, May/June 1999.
Guha R., McCool R., Miller E., 2003. "Semantic Search",
Proceedings of 12th international conference on World
Wide Web, Budapest, Hungary, May 2003.
Hai Zh., 2006. "Semantic component networking: Toward
the synergy of static reuse and dynamic clustering of
resources in the knowledge grid", Oct. 2006, Journal
of Systems and Software, V79, 10, p.1469-82.
Kaiya H., Cai Saeki, Ohnishi A., 2005. "Ontology-based
requirements analysis: lightweight semantic processing
approach", Sept. 2005, Proceedings. Fifth International
Conference on Quality Software (QSIC 2005), p.478
Khaitan S. et al., 2006. "Exploiting Semantic Proximity
for Information Retrieval", available at http://
www.cse.iitb.ac.in/~pb/papers/IJCAI-CLIA-
Exploiting-Semantics.pdf
Kirtland Mary, 1998. "Designing Component-based
Applications", Microsoft Press; Pap/Cdr edition,
December 1998, ISBN 978-0735605237
Larvet Ph., 1994. "Analyse des systèmes, de l'approche
fonctionnelle à l'approche objet", InterEditions, Paris.
Larvet Ph., 2006. "Composing Automatically Web
Services through Semantic Tags", ICSSEA 2006,
International Conference on Software and Systems
Engineering and their Applications, CNAM Paris
(France), December 2006.
Mayfield J. and Finin T., 2003. "Information retrieval on
the semantic web: Integrating inference and retrieval",
Proceedings SIGIR 2003 Semantic Web Workshop.
McGuinness D.L., van Harmelen F., 2004. "OWL Web
Ontology Language", W3C Recommendation 10
February 2004, Editors: Knowledge Systems
Laboratory, Stanford University, Vrije Universiteit,
Amsterdam
Narayanan S., McIlraith S., 2002. "Simulation,
Verification and Automated Composition of Web
Services", Proceedings of the Eleventh International
World Wide Web Conference (WWW-11), pp. 77-88,
May 7-11, 2002, Honolulu, Hawaii, USA.
Patel-Schneider P., Siméon J., 2002. "The Yin/Yang Web:
XML Syntax and RDF Semantics", WWW May 2002,
Honolulu, Hawaii, USA. ACM Public.
Patel-Schneider P., Fensel D., 2002. "Layering the
Semantic Web: Problems and Directions" - The
Semantic Web-ISWC 2002: First International
Semantic Web.
Rumbaugh J., Booch G., Jacobson I., 1999. "The Unified
Modeling Language, Reference Manual", Addison-
Wesley, New York (1999)
Sjachyn M., Beus-Dukic L., 2006. "Semantic component
selection", 5
th
International Conference on Commercial-
off-the-Shelf (COTS)-Based Software Systems.
SEMANTIC APPLICATION DESIGN
55