AN INTEGRATED TOOL FOR SUPPORTING ONTOLOGY DRIVEN
REQUIREMENTS ELICITATION
Motohiro Kitamura, Ryo Hasegawa
Tokyo Institute of Technology, Ookayama 2-12-1-W8-83, Meguro, Tokyo 152-8552, Japan
Haruhiko Kaiya
Shinshu University, Wakasato 4-17-1, Nagano 380-8553, Japan
Motoshi Saeki
Tokyo Institute of Technology, Ookayama 2-12-1-W8-83, Meguro, Tokyo 152-8552, Japan
Keywords:
Requirements Elicitation, Ontology, Requirements Engineering, CASE Tool, Text Mining.
Abstract:
Since requirements analysts do not have sufficient knowledge on a problem domain, i.e. domain knowledge,
the technique how to make up for lacks of domain knowledge is a key issue. This paper proposes the usage of
a domain ontology as domain knowledge during requirements elicitation processes and the technique to create
a domain ontology for a certain problem domain by using text-mining techniques.
1 INTRODUCTION
Knowledge on a problem domain where software is
operated (simply, domain knowledge) plays an impor-
tant role on eliciting from customers and users their
requirements of high quality. For example, to de-
velop e-commerce systems, the knowledge on mar-
keting business processes, supply chain management,
commercial laws, etc. is required as well as knowl-
edge on internet technology. Although requirements
analysts have much knowledge of software technol-
ogy, they may have less domain knowledge. As a re-
sult, lack of domain knowledge allows the analysts
to produce requirements specification of low quality,
e.g. an incomplete requirements specification where
mandatory requirements are lacking. Thus, the tech-
niques to provide domain knowledge for the analysts
during their requirements elicitation, and computer-
ized tools based on these techniques to support the
analysts are necessary.
Kaiya et al. have proposed the methodology
called ORE (Ontology driven Requirements Elicita-
tion) (Kaiya and Saeki, 2006) where domain ontolo-
gies are used to supplement domain knowledge to
requirements analysts during requirement elicitation
processes. However, it mentioned just a methodology
but did not address the issues on how the analyst can
utilize a domain ontology more concretely or on how
a domain ontology of high quality can be constructed
with less human efforts. This paper presents auto-
mated integrated tool for supporting the usage and the
construction of a domain ontology. By using this tool,
lacking requirements and inconsistent ones are incre-
mentally suggested to the requirements analyst and
he evolves a list of the current requirements based on
these suggestions. The tool deduces lacking require-
ments and inconsistency ones by using inference rules
on the domain ontology.
Some studies to extract ontological concepts and
their relationships by applying text-mining techniques
to natural-language specification documents exist
(Goldin and Berry, 1997). We apply this technique
to various kinds of documents related on a problem
domain in order to automate partially the construc-
tion of a domain ontology. We customize and adapt
usual logical structure of ontologies into requirements
elicitation. More concretely, as a result, we adopt va-
rieties of ontological types and their relationships so
that an ontology can represent domain knowledge for
requirements elicitation. Thus a newly devised text-
mining technique fit to our ontological structure is
necessary to achieve the construction of domain on-
tologies from documents.
The rest of this paper is organized as follows. In
the next section, we explain the basic idea and show
the logical structure of domain ontologies. In section
73
Kitamura M., Hasegawa R., Kaiya H. and Saeki M. (2007).
AN INTEGRATED TOOL FOR SUPPORTING ONTOLOGY DRIVEN REQUIREMENTS ELICITATION.
In Proceedings of the Second International Conference on Software and Data Technologies - SE, pages 73-80
DOI: 10.5220/0001330000730080
Copyright
c
SciTePress
1. aaa
2. bbb
3. ccc
require
A requirements document “S” (consists of req. items.)
F
int
: interpretation function
Domain Ontology “O” (thesaurus part only)
Figure 1: Mapping from Requirements to Ontology.
3, we clarify the tool for supporting our ORE method,
i.e. requirements elicitation by using a domain on-
tology. Section 4 presents the text-mining technique
how to create a domain ontology from many docu-
ments of this domain, and it also includes an exper-
imental result on how much effort for ontology cre-
ation could be reduced by applying our technique. In
section 5, we show two case studies to explore the
weakness of our tool. In sections 6 and 7, we discuss
related works and our current conclusions together
with future work, respectively.
2 BASIC IDEA
2.1 Using a Domain Ontology
In this sub section, let’s consider how a requirements
analyst uses a domain ontology for completing re-
quirements elicitation. Suppose that requirements
document initially submitted by a customer are item-
ized as a list. At first, an analyst should map a re-
quirement item (statement) in a document into atomic
concepts of the ontology as shown in Figure 1. In the
figure, the ontology is written in the form of class di-
agrams. For example, the item “bbb” is mapped into
the concepts A and B and an aggregation relation-
ship between them. Thus, the ontology plays a role
of a semantic domain in denotational semantics. The
requirements document may be improved incremen-
tally through the interactions between a requirements
analyst and stakeholders. In this process, logical in-
ference on the ontology suggests to the analyst what
part he should incrementally improve or evolve. In
the figure, although the document S includes the con-
cept A in the item bbb, it does not have the concept C,
which is required by A. The inference resulted from
“C is required by A” and “A is included” suggests to
Concept
quality
function
object
environment
constraint
actor
platform
Relationship
is-a
(generalize)
has-a
(aggregate)
synonym
antonym
contradict
apply
require
perform
2
1
{ordered}
symmetric reflective transitive
ambiguity
Figure 2: Ontology Meta model.
the analyst that a statement having C should be added
to the document S. In our technique, it is important
what kind of relationship like “required by” should
be included in a domain ontology for inference.
2.2 Domain Ontology
As in (Kaiya and Saeki, 2006), our domain ontology
consists of a thesaurus and inference rules. Figure 2
shows the overview of a meta model of the thesaurus
part of our ontologies. As shown in the figure, a meta
model of thesauruses consists of concepts and rela-
tionships among the concepts and it has varies of sub-
classes of “concept” class and “relationship”. For ex-
ample, “object” is a sub class of a concept class and a
relationship “apply” can connect two concepts. Var-
ious types of concepts and relationships in Figure 2
are introduced so as to easily represent the semantics
in software systems, and it leads to the development
of newly devised text-mining technique for creating
our ontologies. The detailed explanations of this meta
model can be found in (Kaiya and Saeki, 2006).
3 REQUIREMENTS
ELICITATION TOOL BASED
ON ORE
Based on the technique mentioned in Figure 1, the
supporting tool for our ORE method must be able
to detect the elements that are incomplete, incorrect,
inconsistent or ambiguous, and suggest them to the
requirements analysts. To achieve this task, the tool
must have powerful reasoning mechanism. We use an
SWI Prolog engine because of its flexibility and inter-
operability to Java. Since in practical environments,
requirements are expressed in natural language, we
use it to represent the requirements. Like IEEE830
standard (IEEE, 1998) and use case templates, there
are some forms or prescribed document structures to
write requirements in natural language, however we
ICSOFT 2007 - International Conference on Software and Data Technologies
74
don’t use them for keeping applicability as wide as
possible. Instead, we use hierarchical itemized sen-
tences which are connected with each other in AND-
OR style. Itemized sentences are widely used in doc-
uments.
In ORE method, the two tasks; 1) developing a
mapping between requirements and concepts and 2)
analyzing requirements by using a domain ontology
should be supported. Figure 3 illustrates a screenshot
of the tool for requirements elicitation following the
ORE method, and we will use it to explain the elic-
itation process. The example problem used here is
a library system and its initial requirements are dis-
played in the left upper area of the window. In the
right upper area, a thesaurus part of the ontology “Li-
brary System” is depicted in class diagram form. Note
that our tool is for Japanese and the examples in this
paper are the result of direct translation from Japanese
into English.
3.1 Mapping between Requirements
and Ontological Concepts
To process the meaning of each sentence in require-
ments symbolically, a set of ontological concepts is
related to each sentence in the requirements list. As
shown in Figure 3, the initial requirements of the
library system consists of three itemized sentences.
Lexical and morphological analysis is automatically
executed on the sentences so as to extract relevant
words from them. We use an automated morpho-
logical and syntax analyzer called “Sen”
1
written
in Java in order to detect morphemes and to iden-
tify their parts of speech (lexical categories such as
nouns, verbs, adjectives, etc.). After filtering out in-
significant morphemes (e.g. “be”-verb, articles, parti-
cles and prepositions) and identifying words and their
parts of speech, the tool finds corresponding ontolog-
ical concepts to each morpheme using a synonym dic-
tionary.
We illustrate the above procedure by using a sen-
tence item # 2 in Figure 3. After morphological analy-
sis, an analyst can get the ve words “books”, “shall”,
“highly”, “available”, and “members” as shown in the
first column “word” of the center table of the figure.
And then the tool constructs automatically two map-
pings from the word “books” to the concept “a book”
and from “members” to “member”, because the con-
cepts having the same labels as these words are in-
cluded in the ontology shown in the right window.
See the columns “word” and “concept1” in the cen-
ter table “Result” of Figure 3. However, the analyst
1
https://sen.dev.java.net/
finds that the word “Books” semantically implies “a
copy of a book” in addition to “a book”, and he or she
tries to modify the mapping from the word “Books”
by clicking the class “a copy of a book” in the right
window of the figure showing the ontology. He or she
clicks the button “set” in the bottom of the right win-
dow, and then the new mapping from “Books” to “a
copy of a book” is added. Finally he or she can have
the mappings 1) “books” object(a book), object(a
copy of a book), 2) “available” quality(availability)
and 3) “members” actor(member).
3.2 Analyzing Requirements by Using
an Ontology
By using concepts and morphemes corresponding to
each requirements item, requirements are improved
incrementally in this task. Our tool detects 1) the con-
cepts to be added into the current requirements and 2)
the concepts to be removed from the current require-
ments, by using inference on the ontology, and ad-
vises suitable actions, e.g. adding a new requirement
or removing the existing one, to the requirements an-
alysts. We define the inference rules as Prolog pro-
grams
2
.
By using the mapping mentioned in the previous
sub section, we will intuitively explain how the infer-
ence rules can find the lacks of mandatory require-
ments, i.e. the requirements to be added. See the Fig-
ure 3 and the ontology of Library domain shown in
Figure 5 (b). In our example, a requirements item #1
is mapped to function(check out). However, no items
are mapped to object(member account) even though
there is a require-relationship between function(check
out) and object(member account) as shown in Figure
5 (b). Our tool detects the missing of “member ac-
count” by the inference on this require-relationship,
and gives an advice to add object(member account)
automatically. Following this advice, an analyst adds
new requirements item about member account, e.g.,
“Each member has his/her own member account”.
The rule A1 shown in Figure 4 is written based on
this idea. In the figure, we illustrate a part of these
rules for producing advices to requirements analysts,
and they can be recursively applied to produce the ad-
vices. The rules are represented with Horn clauses of
Prolog. They can be categorized with respect to their
contribution to characteristics of a good Software Re-
quirements Specification of IEEE 830 (IEEE, 1998),
and for example, rules A1, A2, A3, A4 and A5 for
completeness, and A6 for unambiguity.
Let’s turn back to our example, Figure 3. When
2
http://www.swi-prolog.org
AN INTEGRATED TOOL FOR SUPPORTING ONTOLOGY DRIVEN REQUIREMENTS ELICITATION
75
Figure 3: A Tool for Requirements Elicitation.
Rule A1: If concepts(all) includes concept(A), concepts(all) does not include concept(B) and Ontology() includes
require(concept(A), concept(B)), then our tool gives an advice to add or modify requirements item(s) so that con-
cepts(all) includes concept(B).
Rule A2: If concepts(all) includes object(A), concepts(all) does not include function(B), and Ontology() includes ap-
ply(function(B), object(A)), then our tool gives an advice to add or modify requirements item(s) so that concepts(all)
includes function(B).
Rule A3: If concepts(all) does not include object(A), concepts(all) includes function(B), and Ontology() includes ap-
ply(function(B), object(A)), then our tool gives an advice to add or modify requirements item(s) so that concepts(all)
includes object(A).
Rule A4: If concepts(all) includes environment(A), concepts(all) does not include function(B), and Ontology() includes
perform(environment(A), function(B)), then our tool gives an advice to add or modify requirements item(s) so that
concepts(all) includes function(B).
Rule A5: If concepts(all) does not include environment(A), concepts(all) includes function(B), and Ontology() includes
perform(environment(A), function(B)), then our tool gives an advice to add or modify requirements item(s) so that
concepts(all) includes environment(A).
Rule A6: If morpheme x is identical to morpheme y but mapc(morpheme x) is not identical to mapc(morpheme y),
then our tool gives an advice to rename one of these morphemes.
Abbreviations: concepts(item x): a set of concepts related to a requirements item x. concepts(all): the union of the sets
of concepts related to all requirements items in a requirements list. (
x
concepts(item x)). morphemes(item x): a set
of morphemes related to an item x. mapc(morpheme x): a set of concepts such that the mapping between x and the
concept exists. Ontology(): a set of all relationships and concepts of a domain ontology.
Figure 4: Inference Rules for Producing Advices.
ICSOFT 2007 - International Conference on Software and Data Technologies
76
the analyst clicks a CHECK button located in the bot-
tom of the center area in the figure, the tool starts
inference to detect lacking requirements, inconsis-
tent ones and ambiguous ones. In this example, ac-
cording to the ontology “member” requires the con-
cept “member account” to distinguish members of
the library from non-members, and the rule A1 men-
tioned above suggests that “member account” should
be added. This inference result is shown in the bot-
tom area “attention” in the figure, and he or she can
add the sentence related to “member account” at the
top area “Edit”.
Our tool also tells us the measures of correctness,
consistency, unambiguity and completeness with the
ratio of the detected faults to total number of re-
quirements items. Their calculation technique can be
found in (Kaiya and Saeki, 2006). The left bottom
window of Figure 3 shows the calculation results of
these values. They help an analyst to decide when she
may terminate her elicitation task.
4 A SUPPORTING TOOL FOR
ONTOLOGY CREATION
In this section, we focus on the technique to extract a
thesaurus from Japanese text documents. Basically,
nouns and verbs included in the documents corre-
spond to the object concepts and functions of Figure
2 respectively, and adjectives and adverbs modifying
objects or functions represent the concepts of qual-
ity. Thus the essential parts of our process for on-
tology creation are 1) Word extraction for extracting
from the documents the significant words and terms
that can be considered as ontological concepts, and
2) Relationship extraction for discovering the rela-
tionships among the extracted words and terms (we
simply call them words). After morphological anal-
ysis and dependency analysis, we identify part-of-
speech categories of the meaningful words appearing
the documents such as nouns, verbs, adjectives etc.
These steps can be performed automatically using the
natural-language processing tool Sen, which is used
in section 3. By using part-of-speech information of
words, we classify the words into the types of onto-
logical concepts such as object, function and quality.
For example, “Customer Information” is a noun and
is classified into an object concept.
In the next step, i.e. Word extraction, our tool cal-
culates various measure parameters of the words so
that we can filter out unimportant words from the clas-
sified words. The parameters that we use are based on
word frequency, i.e. the number of times a word ap-
pears in documents, and are shown below.
1. TF (term frequency): the number of times a word
appears in the documents.
2. TF × IDF (term frequency × inverse docu-
ment frequency) : the term frequency of a word
weighted with its importance degree resulting
from the number of the documents it appears.
3. Entropy: logarithmic value of the term frequency
of a word weighted with its entropy.
4. C-value : the term frequency of a word weighted
with its length and its occurrences as a part of
multi-words.
Figure 5 (a) shows an example of the result of word
extraction. As shown in the figure, the words are mea-
sured and they are sorted in descending order of the
measure values. An ontology creator can select the
important words denoting ontological concepts in a
problem domain, by checking boxes on the sorted list
of the measured words. In the example of the figure,
the words “publish”, “lending”, “search”, “print” and
“return” have been manually selected by the creator.
After selecting the words, the creator proceeds to
the step of extracting relationships among the words.
The supporting tool calculates the number of times a
pair of words included in a sentence appears in the
documents, i.e. co-occurrence frequency (CF) of two
words and cosine similarity (CS) of co-occurrence
frequency vectors, in order to find semantically rele-
vant word pairs. After calculating CFs and CSs, pairs
of words whose CF and CS are higher than certain
thresholds are selected as candidates of ontological
relationships. Based on types of words (e.g. object,
function and quality) and dependency structures in the
sentences, the tool suggests the types of the ontolog-
ical relationships. For example, suppose that the CF
of the word u and v is higher, and u and v are “ob-
ject” and “function” respectively. In addition, if u is
an object in grammatical sense and v is its verb, the
tool suggests an “apply” relationship between u and
v. Cosine similarity is useful to detect synonyms, is-a
and has-a relationships. Figure 5 (b) shows an exam-
ple of the detected concepts and their relationship of
Library domain in a class diagram-like form. The tool
users can modify the detected relationships and edit
the diagram to make it more complete and precise as
a domain ontology.
To assess our thesaurus-creation tool, we made an
experiment to measure the effort of worker’s activi-
ties. We had 8 documents on software for making
estimates, whose lengths were from 3 to 23 pages of
A4 paper size and used them as inputs of the tool to
create a thesaurus of “making an estimate” domain.
We had three subjects, and one of them developed a
domain thesaurus manually by referring to the 8 doc-
AN INTEGRATED TOOL FOR SUPPORTING ONTOLOGY DRIVEN REQUIREMENTS ELICITATION
77
Table 1: Comparative Results Between S1 and S2 in Case
Study 1.
Initial S1 S2
Number of requirements
items
14 (1.0) 60 (4.2) 44 (3.1)
Number of mapped con-
cepts
19 (1.0) 39 (2.0) 34 (1.7)
uments. He completed the thesaurus after three days,
and might spent totally more than 24 hours. Two of
our subjects used the tools and they spent 260 and
180 minutes respectively in creating their domain the-
sauruses. They created 432 and 363 relationships be-
tween their extracted words respectively, and 147 of
them were the same. These results can lead us to the
conclusion that our tool significantly helps in efficient
thesaurus-creation, although manual activities still re-
main.
5 CASE STUDIES
In this section, we present experimental results ob-
tained from two case studies of requirements elicita-
tion in order to assess our tool.
5.1 Case Study 1
In our first case study, two skilled students elicited
requirements for “a record management system in a
school”. They did not know the details of this system,
but were its users. One student S1 used our tool to
elicit requirements, while another student S2 did not.
By comparing their results and processes, we discuss
the advantages and disadvantages of our tool for re-
quirements elicitation.
The comparative result is shown in Table 1. The
initial list of requirements that was given to our
subjects consisted of 14 itemized simple sentences.
Through the requirements elicitation processes, S1
and S2 got 60 and 44 itemized simple sentences as
final requirement lists respectively. The second row
in the table expresses how many concepts the final re-
quirements items were mapped to. Although S2 did
not create the mappings from the sentences to the con-
cepts of the thesaurus during his requirements elicita-
tion process, we asked him to make the mapping after
this experiment so as to get this second row of Table
1. For example, the 60 itemized sentences that S1 pro-
duced were mapped into 39 concepts of the thesaurus
during the experiment, while S2 mapped the 44 sen-
tences into 34 concepts after the experiment was fin-
ished. Since each item in the requirements list can
have different amount of its meaning, we calculated
and compared the number of concepts related to each
requirements list. Each number in parentheses in the
table shows the ratio to initial items. For example, the
number of requirements item was extended about 4.2
times of the number of initial requirements items. The
table 1 suggests that our tool can significantly help an
analyst to extend the requirements list, in particular
to find the lacks of requirements, because the number
of items was extended about 4.2 times by S1 but 3.1
times by S2. With respect to the amount of mapped
concepts, the difference is not so significant: 2.0 and
1.7. It can be considered that many items that S1 pro-
duced referred to the same concepts but they were de-
scribed in detail and were further refined than S2.
5.2 Case Study 2
The second case study used a subject unfamiliar to
a problem domain in order to assess our ORE. We
set an example of a POS (Point Of Sales) system for
convenience stores. The subject was a skilled stu-
dent for software design and programming but did not
have knowledge on business processes of convenience
stores or on POS systems. He was given 6 initial re-
quirements and an ontology of POS domain having
43 concepts and 49 relationships, which had been cre-
ated with our tool. Finally he elicited 31 requirements
items (incl. the initial 6 items) after 2 hours. 7 items
of newly added 25 (31-6) items resulted from the the-
saurus part of the ontology (which was indicated in
the right area of the tool screen as shown in Figure 3),
and 4 items from the advices that our tool suggested.
This result that 44% items were suggested with the
ontology showed the useful contribution to require-
ments elicitation by domain non-experts. Since the
size of the ontology was small, the subject did not pay
much attention to the advices derived from the infer-
ence, but rather to the list of the words indicated in the
right area of the tool window so as to get a full view
of the significant domain-specific words.
6 RELATED WORK
In research community of Ontology, many computer-
ized tools for supporting ontology creation have been
developed, and almost of them were diagram editors
with simple syntactic checking mechanisms such as
KAON and Prot
´
eg
´
e where users can input and edit
ontologies visually. Text2Onto of KAON (Cimiano
and Volker, 2005) is a pioneer of the tools having a
text-mining function based on TF × IDF measure so
that words frequently appearing can be extracted from
text documents. In fact, our ontology creation tool
ICSOFT 2007 - International Conference on Software and Data Technologies
78
(a) Word Extraction
(b) Relationship Extraction
Figure 5: Ontology Creation Tool.
AN INTEGRATED TOOL FOR SUPPORTING ONTOLOGY DRIVEN REQUIREMENTS ELICITATION
79
uses the same quantification techniques for word ex-
traction. However, our ontologies have varieties of
types of concepts and of relationships in order to ap-
ply to requirements elicitation, and Text2Onto can-
not classify the extracted concepts and relationships
into these types. OntoLearn (Navigli et al., 2003) uses
WordNet to detect semantic relationships among ex-
tracted words and this technique can be applied in or-
der to make our tool more elaborate. Activity First
Method (AFM) (Mizoguchi et al., 1995) is a method-
ology to extract task ontologies from natural language
texts manually. Unlike ours and Text2Onto, it adopted
the approach based on the occurrences of verbs in the
texts so as to construct the conceptual structures of
tasks corresponding to the verbs. In (Kof, 2004), a
case study to try to extract an ontology from a require-
ments document was presented, and the extracted on-
tology has been used as a formal design model of a
software system to be developed. Its aim of this work
is different from ours because a domain ontology is
not considered as domain knowledge in it.
We could find several studies on the application of
ontologies to requirements engineering. LEL (Lan-
guage Extended Lexicon) (Breitman and Leite, 2003)
is a kind of electronic version of dictionary that can
be used as domain knowledge in requirements elic-
itation processes. Although it includes tags and an-
chors to help analysts fill up domain knowledge, it
has neither methodologies as guidance procedures nor
semantic inference mechanisms. A feature diagram
in Feature Oriented Domain Analysis can be consid-
ered as a kind of a domain thesaurus representation,
and in (Zhang et al., 2005), a technique to analyze
semantic dependencies among requirements by using
features and their dependency relationships was pro-
posed. However, its aim is different from ours.
7 CONCLUSION
The tool that we developed includes two contribu-
tions; the first one is that the inference mechanism
implemented with Prolog helps a requirements ana-
lyst to evolve requirements systematically by taking
account of the semantic aspect of requirements, and
the second one is that the tool supports domain on-
tology creation by using integrated metrics for text-
mining. We partially assess the user-friendliness and
effectiveness of our tool through experiments. How-
ever, our experiment mentioned in section 5 was too
small to argue the generality of the experimental find-
ings.
The quality of requirements elicitation using our
tool greatly depends on the quality of domain on-
tologies. Although we adopted text-mining approach
from the existing documents such as manuals, we
have to improve the approach moreover. Although
our current approach is based on the frequency of
words in documents, frequent words are not always
important in general. Comparing different documents
(Lecceuche, 2000) is one of the ways to complement
the frequency based approach. Another way to cre-
ate an ontology of higher quality is the integration of
many ontologies existing over Internet, which have
been developed by XML, OWL and Ontology com-
munity.
REFERENCES
Breitman, K. and Leite, J. (2003). Ontology as a Require-
ments Engineering Product. In Proc. of 11th IEEE
Requirements Engineering Conference (RE01), pages
309–319.
Cimiano, P. and Volker, J. (2005). Text2onto : A framework
for ontology learning and data-driven change discov-
ery. In Lecture Notes in Computer Science, volume
3513, pages 227–238.
Goldin, L. and Berry, D. (1997). AbstFinder, A Prototype
Natural Language Text Abstraction Finder for Use in
Requirements Elicitation. Automated Software Engi-
neering Journal, 4(4):375 – 412.
IEEE (1998). IEEE Recommended Practice for Software
Requirements Specifications. IEEE Std. 830-1998.
Kaiya, H. and Saeki, M. (2006). Using domain ontology
as domain knowledge for requirements elicitation. In
Proc. of 14th IEEE International Requirements Engi-
neering Conference (RE’06), pages 189–198.
Kof, L. (2004). Natural Language Processing for Require-
ments Engineering: Applicability to Large Require-
ments Documents. In Proc. of the Workshops, 19th
International Conference on Automated Software En-
gineering.
Lecceuche, R. (2000). Finding Comparatively Important
Concepts between Texts. In The Fifteenth IEEE In-
ternational Conference on Automated Software Engi-
neering (ASE’00), pages 55–60, Grenoble, France.
Mizoguchi, R., Ikeda, M., Seta, K., and Vanwelkenhuy-
sen, J. (1995). Ontology for modeling the world from
problem solving perspectives. In IJCAI Workshop on
Basic Ontological Issues in Knowledge Sharing.
Navigli, R., Velardi, P., and Gangemi, A. (2003). Ontology
learning and its application to automated terminology
translation. IEEE Intelligent Systems, 18(1):22–31.
Zhang, W., Mei, H., and Zhao, H. (2005). A Feature-
Oriented Approach to Modeling Requirements De-
pendencies. In Proc. of 13th IEEE International Con-
ference on Requirements Engineering (RE’05), pages
273–284.
ICSOFT 2007 - International Conference on Software and Data Technologies
80