Online First

Classification is carried out to combine similar things into one category and to divide different things. The general result of classification is to arrange concepts and objects in a systematic order (Ko, Seo, & Cho, 2006; Jeong, 1997). A classification system aims to concentrate literature or knowledge of the same or similar subjects into one place, and to provide easy search access. It now covers classification of e-information sources as well.

Classification systems are classified into academic classification, literature classification, technical classification, industrial classification, product classification, encyclopedia classification, information service institution classification, research classification, subject classification, project classification, biological classification, and the like. There is no limit to classification types, especially because the literature classification matches the concept of logical classification and is on the same axis as that of the academic classification theory for academic activities.

The literature classification is for logically and systematically classifying data in libraries depending on their similarity, and is characterized by being more specific and practical in comparison with the academic classification because it is for classifying scientific results. Typical literature classification tables include the DDC (Dewey Decimal Classification), the KDC (Korea Decimal Classification), the UDC (Universal Decimal Classification), the LCC (Library of Congress Classification), and the CC (Colon Classification).

The academic classification, closely related to the library classification, is characterized by its abstract contents, of which the exemplary case is the academic classification system (with 18 categories) of the Korea Research Foundation.

In addition, the technical classification includes science and technology standard classification and industrial technology standard classification. The industrial classification includes the SIC, and the product classification includes the HS, the UNSPSC, and the military supply classification. It is necessary that the academic classification, the literature classification, the industrial classification, and the product classification are specially connected. Classifications can be connected to each other among technologies to which the academic principle is applied, products developed on the basis of those technologies, and the industry to generalize those ideas.

On the other hand, the classification system is divided into enumerative classification and analytic- synthetic (faceted) classification, depending on the format of subject arrangement. Some researchers argue that the analytic-synthetic classification is different from faceted classification. More detailed classification can be achieved through 6 types: enumerative classification, quasi-enumerative classification, quasi-faceted classification, fully fixed faceted classification, quasi-free faceted classification, and free faceted classification.

Enumerative classification predefines symbols about all subject categories, and is a classification system enumerating the symbols with a given rule by means of a top-down approach. The system shown as a symbol is practical, and enables an easy approach to the same category, and higher/lower categories. However, it is difficult to accept new subjects, and requires continuous amendment. The same concept may repeat or appear in different subjects or categories. An exemplary enumerative classification is the LCC.

Faceted classification is a classification system for synthesizing numbers in the final categories by means of predefined rules. This faceted classification analyzes a compound subject into facets in the step of concept and synthesizes them in the steps of language and symbolization, respectively (Jeong, 1997). Providing a variety of auxiliary tables, this faceted classification simplifies the classification system to implement flexible classification, but increases the complexity of tasks for giving classification. An exemplary faceted classification system is the UDC. The DDC and the KDC, exemplary literature classifications, are in the intermediate stage between the enumerative classification system and the faceted classification system.

In addition, classification systems defined according to arrangement structure include list type, tree type (taxonomy), arrangement type (facet), and network type (ontology).

This section relates to the application of classification systems to NDSL (National Discovery for Science Leaders) which is a science and technology information integration service of the KISTI and describes features thereof.

Each different classification system is applied to different types of data and used in different configurations. In particular, although the features of Korea’s journal articles and overseas journal articles are similar, both the DDC classification and the KISTI Standard Classification are used because information provision institutions are different. Accordingly, it is not easy to construct the function of literature reading through subject approach of two types of articles. The journal article, patent, research report, trend and analysis information in the life cycle process of research and development is characterized by a very clear context thereof. That is, it is possible to know the latest research trend and details of precedent research through the trend and analysis information in advance, and patents and articles are produced as a product of the process of research and technology development. Because resultant research reports contribute to spreading research achievements and technology, provision of information for connecting them is very significant in terms of supporting research and development. Therefore, efficient organization and connection services for such information are required. One of the most important tasks to achieve this purpose is the inter-connection of science and technology literature by connecting classification systems.

Korean and overseas researchers have studied how to connect classification systems in order to manage and provide information with each different classification system.

Michael Day, of the Renardus Consortium (2002) tried to connect classification systems for the purpose of subject gateway service among major libraries in 7 European countries. The Renardus Consortium was established for the project of providing a cross-search and reading service of information in the 6 major national libraries of Germany, Denmark, Finland, the Netherlands, Sweden, and the UK. The most characteristic feature of the Renardus service is to provide an integrated subject gateway search service for online academic information which belongs to each library. Mapping classification systems are required among the libraries which are members of the Consortium for the subject gateway search. To this end, Renardus developed a mapping system (using the CARMEN tool) between the DDC classification system and the local classification (LC) system, which was used to build a 1:1 mapping table. When a user carries out a subject gateway search on the Renardus service web-page, the mapping table between the DDC and the LC is used to enable the classified information in the Internet information resources possessed by the libraries in each country to be read. The classification system connection structure of the Renardus system is characterized by unidirectional mapping from the DDC classification to the LC, which results from the feature of the subject gateway.

Relation type	Mathematical Symbols	Example	Description
Fully Equivalent			More or less fully, not necessarily totally congruent
Narrower Equivalent			The potential content of the whole local class is comprised in the DDC class : The Local class is a true subset of the content of the DDC class
Broader Equivalent			the potential content of the whole DDC class is comprised in the local class : The Local class is a true superset of the DDC class
Major Overlap			The local class has a major overlap with the content of the DDC class, related to the full content of the DDC class
Minor Overlap			The local class has a minor overlap with the content of the DDC class, related to the full content of the DDC class. Normally a local class needs to be mapped to several classes.

Hong and Seo (2004) made an analysis of equivalence/non-equivalence between classification systems which may occur in exchanging information between databases to which two classifications are applied, focusing on ‘Science and Technical Standard Classification Codes’ and ‘KISTI Standard Classification (2002)’. His analysis also focused on the classification codes in the field of chemistry. He defined 5 relation types between classification classes for interoperability among classification systems, and stated the issue of non-equivalence when the actual sample classes were tested.

Song and Seol (2006) made an analysis of the feasibility of building the NTIS by using a compatibility table for connecting the Standard Science and the Technical Classification Codes used in the NTIS with the classification systems of research management institutions of each ministry. The analysis revealed that bidirectional interoperability between the NTIS and the classification systems in each research management institution is not very efficient, and it is necessary to consider a new classification system.

Generalizing the above studies, there have been many studies about integration of classification systems by establishing a new classification system, and some studies have even suggested a newly designed classification system. Of course, there has been no actual application of the suggested classification systems yet. This is due to the difficulty of establishing a new standard classification system which satisfies all classification systems for integrating the classification systems, and the amount of time and expenses needed to replace existing classification systems by a new standard classification system, although it could be implemented.

While some overseas studies have suggested a method of connecting classification systems, some studies in Korea have suggested methods of implementation, but no study has suggested a real connection method implemented yet. A mapping table among classification systems is essential for connecting classification systems, and a mapping system for creating the mapping table should be established. The method of connection by means of classification system mapping is superior because it is not needed to completely replace the existing classification systems by a new classification system as in the method of integrating classification systems, but requires much time and expense for classification system mapping. This study was conducted on the assumption that it is better to ensure interconnection of literature by means of a method of connecting a plurality of classification systems, rather than a method of changing the existing classification systems with about 50 million pieces of information of the NDSL into a new classification model, in consideration of expense and efficiency of the aforementioned two methods. The following section describes a method of creating a classification system mapping table and establishing a mapping system.

Day, Koch, and Neuroth (2005) defined the relation type between classification systems, on the assumption of establishing 5 relations: Fully Equivalent, Narrower Equivalent, Broader Equivalent, Major Overlap and Minor Overlap. Hong and Seo (2004) defined 5 semantic equivalent/non-equivalent types: Fully Equivalent, Narrower Equivalent, Broader Equivalent, Major and Minor Overlap and Non-equivalent between two classification classes. Comparing two relation types, Day, Koch, and Neuroth (2005) divides the overlap type into Major Overlap and Minor Overlap depending on which classification between the DDC classification and his own classification includes more subject categories with respect to the overlap relation. However, Hong and Seo (2004) regards the overlap type as one single type. Day, Koch, and Neuroth (2005) does not recognize the non-equivalent type, but Hong and Seo (2004) includes the non-equivalent type category. It seems that the types are slightly different depending on whether the object of the relation type is a classification system or a classification class.

Relation type	Example	Description
Fully Equivalent		The classification class of classification A is semantically equivalent to the classification class of classification B.
Narrower Equivalent		The classification class of classification A is more detailed, and the classification class scope of classification B is wider.
Broader Equivalent		The classification class of classification B is more detailed, and the classification class scope of classification A is wider.
Partially Overlap		The meaning of two classification classes is partially equivalent.
Non-equivalent		Classification classes not related due to a difference in field.

This study gives a definition of 5 relation types of Fully Equivalent, Narrower Equivalent, Broader Equivalent, Major and Minor Overlap and Non-equivalent between classification classes as shown in Table 2 with reference to the above two relation types.

While the 1:1 connection structure between classification systems has the structure shown in Table 1 and achieves mapping, an extension can be implemented to a multi-classification system connection structure which can automatically define the relation type with another classification system. However, in this case mapping is carried out with one classification system among the existing stored classification systems when mapping is further carried out for another classification system. For example, if the classification system A is mapped with the classification system B, and a classification system C is added to be mapped with the classification system B, automatic mapping between the classification system A and the classification system C is implemented according to an extended relation type. Table 2 shows an exemplary relation type of implementing automatic extension between A and C if the relation type (mapping 1) of the classification systems A and B is Fully Equivalent, and the relation type (mapping 2) between classification systems B and C is Fully Equivalent, Narrower Equivalent and Broader Equivalent, respectively.

Mapping 1	Mapping 2	Mapping 3	Description
(Fully Equivalent)	(Fully Equivalent)	(Fully Equivalent)	If both types are Fully Equivalent, the result of automatic mapping is also Fully Equivalent.
	(Narrower Equivalent)	(Narrower Equivalent)	In the relation type including Narrower Equivalent, the result of mapping is also Narrower Equivalent.
	(Broader Equivalent)	(Broader Equivalent)	In the relation type including Broader Equivalent, the result of mapping is also Broader Equivalent.

In this case, if any one mapping is Fully Equivalent, relatively exact equivalence can be implemented for all types other than the mapping type which is Major and Minor Overlap and Non-equivalent.

Mapping 1	Mapping 2	Mapping 3	Description
(Narrower Equivalent)	(Broader Equivalent)	(?)	Mapping according to types Narrower Equivalent and Broader Equivalent should consider human judgment or other elements.

However, if the Narrower Equivalent type is combined with the Broader Equivalent type as shown in the exemplary extension of Table 3, it is necessary to carry out mapping or connecting the classes of the classification system A with those of the classification system C manually or by means of other elements. It seems that an independent process should be provided for the classes with the Non-equivalent type or the Minor and Major Overlap type.

The mapping system for classification systems supports interconnection of science and technology literature by creating mapping tables of the DDC Classification applied to the NDSL science and technology information, the KISTI Standard Classification the INSPEC Classification and the IPC Classification. It provides an interface to allow experts in a specific field and classification experts to define a relation about classification categories of two classification systems finally to create a mapping table for the classification systems. Figure 2 and Figure 3 show how to map two classification systems, composed of classification system search, output from layers of the classification systems, and a component for establishing classification system relation.

Fig. 2. 
Interface of Classification Mapping System

Fig. 3. 
Example of establishing classification system relation

In the interface of Figure 3, the classification relation shown in Figure 4 should be allowed to be defined through the component for establishing classification system relation. Figure 4 shows mapping relation for the items of the KISTI Standard Classification (2005 edition), computers in the INSPEC Classification, and database management systems. MAJ202 in the KISTI Standard Classification and C6160 of the INSPEC Classification achieve semantic full equivalence of Table 1. In this case, MAJ202 of the KISTI Standard Classification is in automatic Broader Equivalent relation with the sub-category of C6160 because of a difference between the structure of the KISTI Standard Classification and that of the INSPEC Classification. However, because the sub-categories of C6160 and 005.75 are not fully equivalent in the INSPEC Classification and the DDC classification, C6160 and 005.75 are not in fully equivalent relation, but in overlap relation. Because of the semantic ambiguity of 005.752, 005.754, 005.755, 005.759 and Other in C6160Z, there is no semantic relation, but it is possible to establish a Broader Equivalent relation. The mapping table shown in Table 5 can be created by using the mapping system for classification systems to establish mapping relation of Figure 4.

Fig. 4. 
Example of Mapping Structure

The mapping table in Table 5 should be designed to enable bi-directional relation setting information for two classification systems to be stored, in order to implement access to one classification system although the other classification system is used.

The mapping system for classification systems designed above is configured in a specialized form by analyzing characteristics of a plurality of classification systems applied to the NDSL. Because most classification systems of the NDSL follow the enumerative classification which defines and lists symbols about all subject categories according to a given rule, system configuration is relatively good. However, if it is a classification system of which the structure is not simple and allows symbol synthesis like the DDC Classification, connection with other classification systems is very complex, and may require much time and expense for mapping. Therefore, this point should be considered. Exceptions and considerations described above that may be encountered in implementing the mapping systems are described below.

• How to establish a relation with subject categories if the two classification systems are of different types.

• How to establish a relation if the two classification systems are of different configurations, such as the enumerative classification and the faceted classification.

• How to handle the information of a created mapping table if it is updated, following amended classification systems.

• How to automatically create a mapping table.

• How to verify the relation established between two classification systems.

• How to automatically extend a category relation with other classification systems which use an existing mapping table.