automatic procedure for building domain specific ... not the case when they face a question like â list all ... In this paper we try to make use of Ontology to ... Ontologies can possess inference functions, ... meaning. 3) Hyponymy: Words that have subset/superset or is-a relation ... There are basically two ways to create an.
Toward enhanced Natural Language Processing to Databases: Building a specific domain Ontology derived from Database Conceptual Model Rajeh Khamis and Safwan Shatnawi University of Bahrain, Applied Studies College {rkhamees,sshatnawi}@asc.uob.bh Abstract . Natural Language Interface to database great success when dealing with simple data NLIDB applications achieve great success when queries written in natural English, however most dealing with simple user requests, however most of NLIDB applications fail when dealing with NLIDB applications fail dramatically when users indirect requests. They could successfully obtain issue indirect or sophisticated requests. One modern the correct results for a question such as “List all approach to enhance NLIDB is using Ontology. students who got grade more than 60” but this was Ontologies are very helpful when used with Natural not the case when they face a question like “ list all Language Processing applications for supporting passed students” considering the success grade is extraction of relevant elements from databases. 60. In this paper we try to make use of Ontology to This paper proposes a framework and a semi help NLIDB applications to correctly understand automatic procedure for building domain specific sophisticated data requests. Ontology can be very Ontology by using data conceptual model and useful in improving NLIDB systems as it allows to general purpose Ontology such as WordNet. The abstract the information and represent it explicitly aim is to help NLIDB understanding and simplifying highlighting the concepts and relations and not the indirect users data requests. words used to describe them. In addition Ontologies can possess inference functions, I. INTRODUCTION allowing more intelligent retrieval. In general using Ontology is considered as a part of semantic Nowadays, the number of nontechnical people approach that aims to use meta data in order to who access a wide range of database has answer user queries, rather than the syntactic dramatically increased therefore, adequate training approach which searches for keywords. Syntactic is needed on the database applications to enable approach suffers from vocabulary inconsistency users smoothly retrieving and manipulating where information object contain relevant information. Such training is not always available information but is not retrieved because it uses and sometimes the database applications are very different words to describe it and its 'opposite' in sophisticated and not designed to be used by such which irrelevant information is retrieved due to users. An ideal communication channel between similarity of words [2]. nontechnical user and computer applications The paper is organized as follows. In section 2 would be natural language, therefore many we discuss three general topics related to this researchers investigated designing Natural work; WordNet, Ontologies and Data Conceptual Language Interface to databases. Natural language Model. A brief description in section 3 discusses interface to databases is a system that allows the some related works. Section 4 explains the user to access information stored in a database by proposed system by introducing the methodology typing requests in natural language [21]. The of the approach and briefly discussing the system then maps each natural language question technical details of the system implementation. to appropriate SQL statements for answering The paper ends up with summary and conclusion. questions or –sometimes-modifying the database structure[22]. In our previous work [1], we have designed and implemented a framework for building Natural Language Interface to Databases (NLIDB). The system and other many similar systems achieve
1 of 9
II.
available [23]. In the following paragraph, few definitions for Ontology are discussed. The first and probably the oldest definition of Ontology was given by Gruber in Mid of Nineties: An Ontology is a formal, explicit specification of a shared Conceptualization [4]. A conceptualization is an abstract model of a phenomenon, concepts, the relations between them and the constraints on their use are explicitly defined. Ideally the Ontology captures knowledge independently of its use and in a way that can be shared universally, but practically different tasks and uses call for different representations of the knowledge in an Ontology [2]. Therefore, an ontology defines a set of representational terms, that typically include concepts and relations among the concepts to describe a target world. Another specific definition is found in [20]: An ontology is the formal specification of the vocabularies of concepts and the relationships among them. 1) Building Ontologies Although research on Ontology has been done since last decade, there is no standard protocol or popular style to build an ontology easily therefore many authors agreed that there is no easy way to build a good ontology[3,7]. An ontology can be constructed in two ways, domain dependent and generic[2]. Domain Ontologies are designed to represent knowledge relevant to a certain domain type, e.g. medical, mechanical etc. While Generic Ontologies can be applied to a variety of domain types . There are basically two ways to create an Ontology. The first and the most obvious one is to build an Ontology from "scratch", i.e. to define classes, relations instants and so on. It is clear that such way is complex and time consuming [16]. The second way is to combine available Ontologies in several forms. The most frequently used forms are: a) Inclusion of one Ontology into another.
THEORETICAL BACKGROUND
A. WordNet WordNet was developed by the Cognitive Science Laboratory at Princeton University and now is considered as one of the largest lexical Ontologies for natural language processing systems [3,5]. It is an online lexical reference system whose design is inspired by current psycholinguistic theories of human lexical memory. English nouns, verbs, adjectives and adverbs are organized into sets of synonyms, each representing one underlying lexical concept. The first goal of WordNet project is the creation of dictionary and thesaurus, which could be used intuitively while the next purpose of WordNet is the support for automatic text analysis and artificial intelligence. WordNet is also useful for determining semantic connections between sets of synonyms, for tracing morphological connections between words [12]. WordNet stores the semantic relationships between the words that include: 1) Synonyms: Words with same or almost similar meaning. 2) Antonymy: Words that are opposite in meaning. 3) Hyponymy: Words that have subset/superset or is-a relation between them and 4) Meronymy: Words that have part of relation between them. In our work we used WordNet because it is a generic ontology which provides a large number of concepts that may contribute to our system. B. Ontology The term Ontology has been borrowed from philosophy where the ontology is a systematic account of existence and introduced to the computer research field during 1990’s by several AI research communities to describe a proper representation of the world in a program code. Recently it has been used in several other IT fields such as intelligent information integration, information retrieval on the Internet, and knowledge management [2,4]. Because many authors propose different ontology definitions, the field of Ontology is still a very volatile field [16,18] and researchers are still trying to look for ways to understand, create and manipulate Ontologies. On the other hand, another group of authors claimed that Ontology technology is nowadays mature enough whereas many methodologies, tools and languages are already
b) Restriction: An Ontology is applied to a restricted subset of what it was originally intended for. c) Refinement: General Ontologies sometimes require refinement in order to be applicable to specific needs[2]. In the proposed system we implement this form of Ontologies development scheme. According to [7] the following points summarize the steps needed to develop an ontology: a) Identify the system concepts
2 of 9
III.
b) identify the main components of concepts, i.e. "part of" relation-.
RELATED WORKS
Many applications have been developed around the idea of Ontology. For example [19] proposed an approach to reverse engineering of relational databases to ontologies based on the idea that semantics of a relational database can be extracted by analyzing HTML forms, without an explicit analysis of relational schema, tuples and user queries. [20] shows that Ontologies can play an important role for heterogeneous database integration. In work [27] the authors concluded that ontologies are semantically richer than database conceptual schemas, and thus closer to the user’s cognitive model. However existing conceptual data models can be useful for building ontologies, since they are formal documents that have been designed to capture the specialist’s view of some aspect of the real world. The work [16] found that existing conceptual data models can be used to create rough ontologies, in contrast existing ontologies can be used to generate conceptual schemas, with or without the aid of an expert modeler. According to [28] the process of ontology construction from relational databases involves analyzing the database schemas in detail. This analysis helps in determining the main entities that may be supported by an environmental database. It may also help in gathering information spread in different tables and determine relationships between different entities identified. It may be noted that the process of mapping ontological elements to the data items in an environmental database can be performed simultaneously [16].
c) identify the main attributes of concepts– could be candidates concepts. d) Connect the concepts with ''is a'' relation to build an initial hierarchy. e) Add detailed term definitions including synonyms. It is important to note that defining terms in detail is not recommended until finalizing the "is a" hierarchy. A carefully designed "is a" hierarchy yields solid context to define the terms [7]. In general, creating Ontology is still a very difficult task. Firstly, most ontologies are created from scratch. Secondly, it is hard to find suitable tool which supports wanted ontology languages or provides possibility to use wanted methodologies for ontology creation. Moreover, ontology integration is not solved issue yet and finally it is difficult to reuse other ontologies or knowledge [16]. C. Data Conceptual Model The use of a conceptual model to describe relational data sources has been proved to be extremely useful to overcome many important data access problems [15]. Conceptual models are descriptions of the organizational context for which a system is developed, and are used to help understanding this context. It might be used to facilitate the design and implementation of an information system or to evaluate the fit between an organization's needs and the business models embedded within an enterprise application software package [16,36]. Conceptual schemas are built to organize what is going to be stored in a database, and then are used to document it. The basic components of ER diagram are[16, 37]: 1) Entities. An entity is a distinctly identified phenomenon that can be classified into entity classes. 2) Relationships. A relationship is an association among entities that can be classified into relationship classes. 3) Attributes and data values. A value is used to give value to a property of an entity or relationship. An attribute is a function which maps from an entity class or relationship class to a value class (group of values); thus the property of an entity or a relationship can be expressed by an attribute-value pair.
IV.
THE PROPOSED SYSTEM
In the following section we discuss the methodology we follow in order to develop the system then a brief description is presented about the natural language processing techniques applied in this system. As shown in the system architecture illustrated in figure 1 the domain Ontology is built first by using the system conceptual model –the ER diagram, general purpose Ontology –WordNet, and domain expert feedback. The generated Ontology is then used to trace, understand and simplify indirect queries asked by user. These simple queries are then passed to NLSQL module which responsible on forming the SQL statements that answer the user question. Finally the database system executes the SQL statement and shows the correct results.
3 of 9
In the following subsections we will explain the proposed approach implemented in the system that extracts the domain Ontology.
diagram) to build an initial ontology. In case the conceptual model is not available, it could be automatically extracted from the existing relational database by using reverse engineering principles and some CASE tools. Figure 2 shows a sample of main components extracted from ER diagram – sketched by using Chen notation. This phase aims to build the basic taxonomy of the ontology.
A. building ontology for a specific domain The process we propose for developing specific domain ontology consists from four phases: 1) Phase 1: Extracting the main concepts of the conceptual model: This phase includes extracting main concepts, attributes and relations from conceptual model (ER
Expert Wordnet Ontology
Ontology
User Input Processing
User
Initial Domain Ontology
Reverse Engineering
Simple Natural Language Query
Natural Language Query
Processed Natural Language Query
Domain
Entity relationship for Domain database
Simple Natural Language Query
NLSQL
SQL Statement
Database
Result Figure 1. The system architecture
Registers
course
Course title
Course type
Student
Name
Figure 2. main components extracted from ER diagram
4 of 9
Gender
the main entity in the conceptual model is equivalent to "Concept" in Ontology Figure 3 shows a graphical representation for the ontology after adding WordNet entries and applying mapping rules. 4) Phase 4: Refining the Ontology The domain experts who are knowledgeable about the application domain are responsible on eliminating unnecessary and irrelative concepts, terms and relations, and modifying the ontology to properly describe the domain. Now we have ready to use specific domain ontology that could be used with NLIDB systems. By virtue of the developed Ontology an indirect query such as “List all passed students” will be analyzed and rephrased to “List all student who got grade greater than 60” whereas this description is exist in the Ontology to describe (PASS) word. Such simple query is analyzable by most of NLIDB applications. Suppose that someone asked about the “name of the boy who meet a specific criteria”. By using the Ontology
2) Phase 2: Adding concepts descriptions, synonyms, hyponyms, and semantic meaning The general lexical Ontology "WordNet" is used to enrich the initial ontology, by adding synonyms, hyponyms, and semantic meaning, descriptions, and discovering more relations among main concepts, sub concepts and attributes. In this phase the system also reviews the data inside the database to extract the domain value for attributes if possible. The domain value is a set of specific values that all attribute value must belong to. For example the domain value for attribute gender is {male, female). Extracting domain value is not applicable for some attribute like (name). Synonyms and descriptions of domain values are also obtained. 3) Phase 3: Applying transformation rules from ER diagram to Ontology In this phase the proposed approach considers and applies the mapping (transformation) rules shown in table 1 to extract the initial domain ontology out of the conceptual model. For example
. Table 1 transformation rules from ER to Ontology DB Conceptual model
Domain Ontology
Example
Entity Attribute Relation between entity and attribute Relation between domain value and attribute Relation among entities
Main concept Sub concept Entity 'has a' attribute
Student Gender Student has a gender
Domain value member 'is an' attribute Entity1 'relation' Entity2 (Non hierarchy relation)
Female is a gender
register
course Has a Course title
Student register in a course
Student Has a
Has a
Course type
Name
gender is a
Male (syn: man, boy, Mr)
Figure 3. The Ontology after adding WordNet entries
5 of 9
Has a
is a Female (syn: lady, girl, Ms)
Major PK
Department
major_id
PK
Dept_no
Belongs to FK1
dept_no
Dname college_id
Enrolls in Student
FK1
fname initial lname major_id
transcript
has
stno course_id grade year semester
Figure 4. ER diagram for the studied system
table shown in table 4 keeps track about all “is-a” relations while the last table illustrated in table 5 stores all synonyms for all concepts, relationships, and attributes in the domain. 2) Building Language State Table The Language State Table (LST) is used to uncover ambiguous terms and simplify the natural language understanding. All terms that describe the entities and semantic concepts in the domain are used to construct the LST which contains all words in the Ontology tables and the action to be taken with each word as generally shown in table 6. The following paragraph discusses how LST is implemented in the context of natural language processing. The first row of the LST contains all words in the Ontology. The user query is divided into separated words. Starting from state 0, each word would have si state; which is either next state or final state (-99). In the latter case the action entry is inserted in the user request instead of the original word. Table 7 shows an example of some words in the LST which was derived from the Ontology being studied. For example if the user query contains “student passed CSA113” then the query is divided into three individual words “[student] [passed] [CSA113]”.
illustrated in figure 3, the system will recognize that the intended entity here is the student, because according to the Ontology rules shown in figure 3" boy is a synonym of male male is a gender student has a gender B. System Implementation We consider part of education and registration system for our study, as illustrated the ER diagram shown in figure 4.In the following subsection we describe the system implementation which has been implemented by using relational database. The First subsection shows how Ontology records are stored in four different tables. The second subsection clarifies the Language State Table (LST) which has been developed to simplify and interpret NL queries. Then, the user input processing is discussed and finally the last subsection explains how NLSQL module works. 1) Ontology Representation: The system maintains all domain Ontology in four tables. The first table shown in table 2 contains information about the entities, its description and the super class –parent- entity. Each entity or relationship has a record that states the entity name, and super class of this entity. For example the third record “major” has a super class department which means that the major is part of department entity. The second table shown in table 3 contains all “has a” relationships to describe the relations between each entity and its attribute. The third
6 of 9
Table 2: WordNet entity relationships description term_id
Term
super
Description
1 2 3
student instructor major
0 0 8
4 5 6 7 8 9
transcript course study teach department enrol
0 2 0 0 0 0
a learner who is enrolled in an educational institution a person whose occupation is teaching the principal field of study of a student at a university a reproduction of a written record education imparted in a series of lessons or meetings student in course taught by instructor instructor impart knowledge to student specialized division of university student registries in course Table 3: “Has-a” description
has_id
term_id
Attribute
Description
1 2 3 4
1 1 1 1
Name Id Address Gender
5 6 7 8 9 10 11
1 1 1 1 3 3 3
Major department Advisor Study Student Instructor Name
a language unit by which a person or thing is known a numeral or string of numerals that is used for identification written directions for finding some location the properties that distinguish organisms on the basis of their reproductive roles the principal field of study of a student at a university a specialized division of a large organization an expert who gives advice follow a course of study a learner who is enrolled in an educational institution a person whose occupation is teaching a language unit by which a person or thing is known
isa_id
Term_id
isa_attribute
Description
1 2 23
4 4 5
female male elective course
a person who belongs to the sex that can have babies a person who belongs to the sex that cannot have babies a course that the student can select from among alternatives
Table 4: “is a” description
Table 6: Synonyms Table 5Language state table
syn_id
word_i d
synonym
1 2
19 19
3 4 5 6 7
19 19 19 21 21
desertion leaving behind leaving rejection neglect lower demean
State
Student
first
name
major
passed
fail
register
Action
0 1 2 3
3 -99 -99 -99
51 -99 -99 -99
52 -99 -99 -99
30 -99 -99 -99
2 -99 -99 -99
1 -99 -99 -99
6 -99 -99 -99
Got Mark =60 in student
word
word
word
word
word
word
word
Action
State
0 1 2
S1* S2 S3 Si Sj Si Sj Si Sj Si Sj Sj Si Sj Si Sj Si Sj Sj Si Sj Sn is the next state; Ai is the action to be taken
Table 7: Language state table - example
7 of 9
A1 A2 A3
be easily substituted by external modules for benchmarking and performance measurement. The following example shown in figure 5 shows how the system modules work to convert the natural language user input to SQL statement.
The LST value for the first word "student" in state 0 is 3. This means that the pointer will go to row 4 (state 3). In row 4 all entries are -99 which means that it reaches to the final state, as a result the system will return the Action exist in Action column. In the “student” case the action is “student” which means that student is a simple word and no further explanation is exist in the ontology . The case is different if we consider word “passed” since the action found there is “got mark >=60”. As a result the user input will be rephrased to be “[student][got mark >=60 in][CSA113]”. 3) User Input processing The user enters the query using natural language, then the system tokenizes the input and sends it to the LST as an array of words to take the appropriate action for each word as described in the previous section. During this stage the user input may be altered to reflect the semantic meaning of the user words. After that the modified input is submitted to the SQL generating module to convert the simple natural language query to SQL statement. The aim of the modules described above is to convert the complex requests into simple ones. Simple request is the request which do not imply domain semantic such as “distinguished students”. All domain semantic information must be converted into simple format like “student who has GPA >=95” for example. This work is done by elaborating the domain ontology to resolve indirect domain semantic words into simple attribute syntax words. 4) NLSQL module. The produced –simplified- natural language query is processed in this module. For further explanation, detailed technical description about this module can be found in our previous work [1]. First the user input is classified into two parts; required fields and conditions. The required fields part is processed to retrieve the database fields and the tables it belongs to, while the conditions part processing phase identifies the operators to be used and the condition value. After identifying the required fields and conditions, the module finalizes the SQL statement by adding join conditions in case the required fields come from two or more tables, adding any aggregate functions required to group the data; and finally the module executes the generated SQL to retrieve the results and show them to the user. In conclusion, this module is able to convert simple natural language input into SQL statements.It is also domain independent that can
V.
SUMMARY AND CONCLUSION
We have proposed a framework that uses general purpose resources like WordNet along with very specific domain artifact like data conceptual model. The aim is to develop specific domain Ontology to be used in NLIDB applications in order to understand and interpret indirect user queries and convert them into simple and semantic meaning free requests. Such simple requests could be directly processed by existing Natural Language Applications . Our major focus was not to reinvent the wheel and avoid building ontology from scratch by make benefit from existing public resources such as WordNet and domain resources such as conceptual model in addition to feedback from domain expert to develop specific domain Ontology. It was found that database conceptual model may be considered as an excellent starting point to start constructing Ontology by creating the initial taxonomy or say the Ontology backbone. WordNet and domain experts carry out totally different functionalities. WordNet played a major role in enriching an initial Ontology and make it bigger and more comprehensive while the role of the domain expert is to refine and remove the unnecessary entries -or make the ontology smaller. The produced Ontology is supposed to be light and serve a specific domain. Ontologies are very helpful when used with Natural Language Processing applications for supporting extraction of relevant elements from databases. Therefore it is expected that a large variety of applications that either utilize Ontologies for improved knowledge management or retrieval, or assist in creating Ontologies more easily and generically, will be emerging in the near future. An interesting future direction might be comparing between two different NLIDB systems for the same domain by using the same specific domain Ontology. Such step might be useful to check and estimate the quality of NLIDB applications. The system is still under development. Study of actual use of this system for specific domains will be part of continuing and future work.
User Query : list all students passed CSA113
8 of 9
[11] . David Vallet, Miriam Fernández, Pablo Castells,”An Ontology-Based Information Retrieval Model”, ESWC 2005: 455-470 [12] Václav Snásel, Pavel Moravec, Jaroslav Pokorný “Using BFA with wordnet ontology based model for web retrieval”, SITIS,, 2005, P 254-259 [13] Derek H. Sleeman, Stephen Potter, David Robertson, W. Marco Schorlemmer: Ontology Extraction for Distributed Environments. Knowledge Transformation for the Semantic Web 2003: 80-91 [14] B. Yildiz, S. Miksch, Motivating Ontology Driven Information Extraction [15] L. Lubyte and S. Tessaris. “Extracting Ontologies from Relational Databases”. KRDB Research Centre Technical Report KRDB07-4, Faculty of Computer Science, Free University of Bozen-Bolzano, Italy. March 28, 2007 [16] Justas Trinkunas, Olegas Vasilecas, “Building Ontologies from Relational Databases Using Reverse Engineering Methods”, International Conference on Computer Systems and Technologies – CompSysTech 2007, vol II, P 6-16.6 [17] Michal Laclavik, “RDB2Onto: Approach for creating semantic metadata from relational database data”; In INFORMATICS 2007 : proceedings of the ninth international conference on informatics. Bratislava : Slovak Society for Applied Cybernetics and Informatics, 2007. ISBN 978-80-969243-7-0, s. 113-116. [18] Man Li Xiao-Yong Du Shan Wang ,“ Learning ontology from relational database”, Machine Learning and Cybernetics, 2005. Proceedings of 2005 International Conference on, Vol 6 2005, P 3410-3415 Vol. 6 [19] 19. Irina Astrova,”Reverse Engineering of Relational Databases to Ontologies”, ESWS 2004},p327-341 [20] Dejing Dou and Paea LePendu,”Ontology-based integration for relational databases”,SAC 2006,p 461466. [21] I. Androutsopoulos, G.D. Ritchie, and P.Thanisch, "Natural Language Interfaces to Databases – An Introduction", Journal of Natural Language Engineering, v. 1(1), Cambridge University Press, pp. 29-81, 1995.. [22] N. Omar, "Heuristics-based Entity-Relationship modeling through Natural Language Processing," PhD Thesis, University of Ulster, 2004. [23] O. Corcho et al. Methodologies, tools and languages for building ontologies. Where is their meeting point? / Data & Knowledge Engineering 46 (2003), pp. 41–64 [24] V. R. Benjamins, J. Contreras, O. Corcho, and A. GَmezPérez. Six Challenges for the Semantic Web. The VLDB Journal, 1, Issue 3„ 2004.. [25] Shanks, G., Tansley E.,Webber R. Using ontology to validate conceptual models. COMMUNICATIONS OF THE ACM October 2003/Vol. 46. No. 10, pp. 85-89 [26] Chen P. P. The entity-relationship model: Towards a unified view of data. ACM Transactions on Database Systems, 1(1), 1976, pp. 471-522. [27] Foncesa F., Davis C., Camara G. Bridging Ontologies and Conceptual Schemas in Geographic Applications Development, Geoinformatica: 7(4), 2003, pp. 355-378 [28] Kashyap V. Design and creation of ontologies for environmental information retrieval. 12th Workshop on Knowledge Acquisition, Modeling and Management (KAW'99), Banff, Canada, October 1999.
User input module : Input : list all students passed CSA113 Output: list all students got mark >=60 in CSA113 NLSQL module : Phase1: attributes and conditions input: list all students got mark >=60 in CSA113 Output: select stno, name, major … from transcripts , students where mark >=60 and course CSA113 Phase 2: finalize SQL input: list stno, name, major … from transcripts , students where mark >=60 and course CSA113 Output: select stno, name, major … from transcripts , students where mark >=60 and course CSA113 and students.stno =transcripts.stno Figure 5. Natural Language to SQL example
VI.
REFERENCES
[1]
(In Press) S. Shatnawi, R. Khamis, "An Approach for developing Natural Language Interface to databases using Data Synonyms Tree and Syntax State Table", International Joint Conferences on Computer, Information, and Systems Sciences, and Engineering (CISSE 09), 4-12 December, 2009, IEEE, Connecticut, USA [2] N. Moriya, “The specification of a shared conceptualization”, Review Document, IQlue 2006. [3] Haitao G., “Semantic Documents for web ontology building”, PHD Thesis, Department of Electrical Engineering & Computer Science, University of California, Irvine, 2005. [4] Smith B., Guid to the philosophy of computing and information 1st ed, Oxford:Blackwell, 2003, 155-166. [5] Christiane F., Wordnet: An Electronic Lexical Database, Cambridge, MA:The MIT press,1998. [6] Vasudeva Varma, "Building Large Scale Ontology Networks," lec, pp.121, Language Engineering Conference (LEC'02), 2002 [7] Mizoguchi R. 2004. Tutorial on ontological engineering: part 3: Advanced course of ontological engineering. New Gen. Comput. 22, 2 (Jan. 2004), 198-220. [8] S. Decker, M. Erdmann, D. Fensel, and R. Studer: Ontobroker: Ontology based Access to Distributed and Semi-Structured Information. In: R. Meersman et al. (eds.), Database Semantics, Semantic Issues in Multimedia Systems, Kluwer Academic Publisher, Boston, 351-369,1999 [9] Paul Buitelaar, Philipp Cimiano, Bernardo Magnini Ontology Learning from Text: An Overview In: Paul Buitelaar, Philipp Cimiano, Bernardo Magnini (eds.) Ontology Learning from Text: Methods, Evaluation and Applications Frontiers in Artificial Intelligence and Applications Series, Vol. 123, IOS Press, July 2005 [10] . Kouji Kozaki and Takeru Hirota and Hiroko Kou and Mamoru Ohta and Riichiro Mizoguchi,”Viewpoint Management for Multi-Perspective issues of Ontologies”,8th International Semantic Web Conference (ISWC2009),2009.
9 of 9