Development of Information Navigation Service Based on Classification Schemes
Pavel Shapkin
MEPhI, Kashirskoe Shosse, 31, Moscow, Russia
Alexander Shapkin
VINITI, Usievicha, 20, Moscow, Russia
Keywords: Classification schemes, classification scheme mapping, information search, XML web services, Semantic
Abstract: Internet allows accessing large document databases contained in different information centres across the
world. Each database has its own search engine which is based on an index or classification scheme. Prob-
lems occur when a user tries to search different databases at once: different databases use different classifi-
cation schemes. This article describes a classification scheme mapping service which is useful in integration
of different databases in one search engine.
Many information centers in the world are process-
ing scientific information. As an example of these
centers we can consider CAS, BIOSYS, Medline,
ISI etc. In Russia the All-Russian Scientific and
Technical Information Institute of Russian Academy
of Sciences (VINITI) is such center. Each centre
contains large amount of scientific information in
form of document databases. Each centre creates an
index of the documents based on a classification
scheme, e. g. Universal Decimal Classification
(UDC) or Mathematics Subject Classification
(MSC). By means of this classification scheme users
can search documents in database (Batty, 1998;
Gilyarevsky, 1971). But if a search involves differ-
ent databases simultaneously, problems occur: dif-
ferent information centers use different classification
Nevertheless problem of searching different da-
tabases at once becomes more and more actual be-
cause of expansion of Internet, which connects dif-
ferent information centers in one worldwide network
(Clarke, 2000). In order to give users an opportunity
to search different document databases at once we
need a public resource which can provide a service
for converting concepts from one classification
scheme to another. It is important that user has not to
be familiar with distinctions in indexing and subject
representation in different information centers.
It is sometimes difficult to define concordances
between different classification schemes. This work
must be conducted by experts. That's why mapping
service can be separated in form of an independent
resource and used as a component of meta-search
engine. This service has to be maintained by organi-
zation which keeps track on changes in classification
schemes and uses experts to define concordances.
Using a mapping service between different clas-
sification schemes user has to know only one classi-
fication scheme. An attempt of building such a ser-
vice is carried out in VINITI. VINITI uses different
classification schemes during the processing of in-
coming document flow. Thus a large amount of
knowledge about concordance of classifications is
accumulated. Furthermore, VINITI is a member of
UDC Consortium, and is responsible for maintaining
Russian version of this classification.
Shapkin P. and Shapkin A. (2007).
SOFTWARE TOOLS FOR NAVIGATION IN DOCUMENT DATABASES - Development of Information Navigation Service Based on Classification
In Proceedings of the Third International Conference on Web Information Systems and Technologies - Web Interfaces and Applications, pages 455-458
DOI: 10.5220/0001285804550458
System of Classification Schemes (SCS) includes
classifications that are used in processing of incom-
ing flow of heterogeneous scientific information.
System allows maintaining a set of classifications
which have complicated structure and also compar-
ing them.
Atomic elements of classifications are called ru-
brics. Classification is a hierarchical structure of
rubrics based on “parent-child” relation among ru-
brics. Each rubric has a code — unique identifier
within the bounds of a classification.
The “parent-child” relation is main but not
unique relation between rubrics. Often it is needed to
represent more complicated interrelations of con-
cepts which are beyond strict hierarchical scheme.
That’s why simple hierarchical model of classifica-
tion scheme gets extended in SCS through introduc-
tion of direct links and other concepts. Furthermore
SCS has means for classification scheme mapping.
Detailed description of these concepts is given be-
2.1 Rubric Properties
Each rubric has some backbone properties and a set
of optional properties which can vary according to
the type of classification scheme.
Backbone properties are rubric code and code of
parent rubric. Within the bounds of one classifica-
tion scheme rubric codes are unique. Each rubric
must have one parent. The only exceptions are root
rubrics of classifications, which have no parents.
Other common property of rubrics is its title
which can be given in different forms: full or short
and on different languages.
Some classification schemes support descriptors
(or keywords): each rubric can be concerned with a
list of keywords. This approach simplifies search of
Each classification scheme changes in time due
to evolution of subject representation in concerned
domain. Formally these changes mean addition,
modification or removal of rubrics.
To maintain lifecycle of a rubric, special proper-
ties are introduced:
date of creation;
date of exclusion from classification scheme;
current status, or lifecycle stage.
Besides the “parent-child” relation direct links
between rubrics within a classification scheme are
allowed. These links represent references like “see
also”, “reference from”, “instead of” etc.
2.2 Classification Scheme Mapping
Comparison of classification schemes allows build-
ing links between rubrics contained in different clas-
sification schemes, in other words, mappings be-
tween classification schemes.
Examine two classification schemes R and Q.
Each of them is a finite set of rubrics:
R = {r
, r
, … r
} where [R] is cardinality of R,
Q = {q
, q
, … q
} where [Q] is cardinality
of Q.
Mapping of rubric r
(source rubric) from classi-
fication scheme R to classification scheme Q (target
scheme) is a set of pairs (o
, p
) where o
is an opera-
tor and p
is nonempty set of rubrics from Q. Opera-
tors define meaning of relation, e. g. “includes” or
“is equivalent to”.
Consider an abstract example of mapping a ru-
bric r
from R to Q, (operators are underlined):
includes q
and q
is included in q
, q
, q
Thus mapping of entire classification scheme R
to classification scheme Q is defined as a set of
mappings of all its rubrics to Q:
R Q = {r
Q where i=1, 2, … [R]}
As follows from the above, mapping has a direc-
tion — from source classification scheme to target
classification scheme and mapping is one-to-many
relation: it contains one source rubric and many tar-
get rubrics.
Initially mappings are created by experts. But
expert work is expensive whereas system contains
more than 20 classification schemes. Experts cannot
create mappings between every pair of classifica-
tions. That’s why some mappings are computed
automatically from expert mappings, these are in-
verse and transitive mappings.
Consider an expert mapping M of classification
scheme R to classification scheme Q. Inverse map-
ping for M is mapping of Q to R, computed from M.
That is if rubric r
is mapped to q
with operator op
in M, then is inverse mapping q
is mapped to r
with inverse operator. Each operator has correspond-
WEBIST 2007 - International Conference on Web Information Systems and Technologies
ing inverse operator, e. g. “is included in” is inverse
of “includes”, etc.
Transitive mapping is a union of a chain of ex-
pert mappings. Suppose there exist expert or inverse
mappings between schemes R and Q, and between Q
and P. Then a transitive mapping of R to P can be
computed from these two mappings. During compu-
tation of transitive mapping compositions of map-
ping operators are used. For example, if rubric r
equivalent to q
, and q
includes p
, then r
. Thus, “includes” is composition of “is equivalent
to” and “includes”.
Introduction of transitive and inverse mapping
gives an opportunity to construct mapping between
virtually any pair of classification schemes with
minimal cost: experts need to build only the basic
mappings, all other mappings will be evaluated “on
the fly”.
2.3 Quantitative Characteristics of
Classification Scheme System
At present system contains 25 classification
schemes, such as Universal Decimal Classification
(UDC), International Patent Classification (IPC),
AMS Mathematics Subject Classification (MSC),
Library and Information Science Abstracts (LISA),
VINITI Classification Index and others. Rubrics
count in a scheme varies from 200 to 67000. Total
number of rubrics contained in system is about 1
Main list of descriptors contains 436000 key-
words and phrases, number of keywords bound to a
rubric varies from 1 to 150.
Inner links between rubrics are used in 11 classi-
fication schemes; mean part of rubrics connected
with direct links amounts to 14% of total rubric
count in classification scheme.
There are 19 expert mappings between separate
classification schemes.
SCS has a web-interface, which is an ASP.NET-
based application (Payne, 2001); data is represented
in XML format (Bean, 2003). The main part the ap-
plication is a set of two web-services:
metadata service;
mapping service.
By means of the metadata service users can re-
ceive information about classification schemes
which are available in the system. One can get a list
of available schemes, lists of their properties and
information on available mappings.
Mapping service allows users to get mapping of
a rubric from one classification scheme to another.
Both services allow access using SOAP, HTTP
GET or HTTP POST protocols.
Because implementation is based on an XML
web-service, it can be accessed from any type of
applications, e. g. from client application written on
any programming language, from a web-site or even
from AJAX-based web page. XML data can be
transformed using XSL (Holzner, 2001).
Along with web-service a simple HTML inter-
face was created. It uses AJAX to utilize metadata
service in order to obtain information about avail-
able classification schemes and their mappings. User
can choose source and target classification scheme
from a list, enter code of source rubric and then re-
ceive the list of rubrics on which source rubric is
mapped. An experimental test page of the service is
located at “
Let us examine an example of utilizing classification
scheme mapping service within an information por-
tal. Consider an internet resource containing a large
database of publications, e. g. of dissertations. Each
dissertation contains a Universal Decimal Classifica-
tion (UDC) index, which determines the area to
which this dissertation belongs. Suppose that portal
implements a new service that enables users to get a
list of patents whose area is close to the area of cho-
sen dissertation. Patents can be obtained from a pub-
licly available service like esp@cenet. The problem
is that patents are indexed with International Patent
Classification (IPC). That’s where the classification
scheme mapping service appears. To get a list of
patents it is needed first to send request containing
the UDC index of chosen dissertation to mapping
service. As a response portal system will receive a
list of IPC indexes related to this UDC index. Then
portal system can use this list in order to retrieve
required patents from patent database. Interaction
diagram for this example is shown on fig. 1
Service Based on Classification Schemes
Figure 1: Classification scheme mapping service utiliza-
tion example.
The main idea of Semantic Web is to make informa-
tion contained in web resources suitable not only for
use by people but also for processing by machines.
Instead of using HTML, which describes only the
representation of data, Semantic Web languages
allow to describe semantics of data explicitly.
One of Semantic Web principles is decentraliza-
tion of data and centralization of metadata. It means
that public metadata resources are needed. Classifi-
cation schemes are perfect examples of metadata,
thus service for accessing classification schemes and
their mappings can act as a metadata provider. In
order to integrate classification scheme service in
Semantic Web it may be necessary to bring XML
data representation format in correspondence with
standards of Semantic Web. It implies using lan-
guages like RDF (Resource Description Framework,
see W3C RDF Primer) and OWL (Ontology Web
Language, see W3C OWL Guide).
The aim of RDF is to standardize format of de-
scribing metadata used in web resources. Main con-
struction of RDF is triplet “object-attribute-value”. It
can be written as A(O, V) which means “object O
has an attribute A with value V”. Attributes are often
called properties or relations, and objects are also
called entities. Each element in the triplet can be
specified with its Uniform Resource Identifier (URI)
— global unique identifier. Triplets can be nested
and thus form a graph.
An example of graph describing a rubric from
VINITI Classification Index is shown on fig. 2. Enti-
ties are represented by ovals, values — by boxes and
relations are represented by arrows. Identifiers are
shown within the objects; types of objects are shown
near them. Entities’ and relations’ URIs are shown.
Use of URIs guarantees that, e. g. Child relation
will be recognized as the parent-child relation used
in VINITI Classification Index.
Figure 2: RDF representation graph for VINITI Classifica-
tion Index rubric.
Classification scheme system can act as a source of
information about different classification schemes
for scientific information. Ability of mapping differ-
ent classification schemes allows to use this system
for purposes of integration document databases of
different information centers.
Payne, C., 2001. Teach Yourself ASP.NET in 21 Days,
Bean, J., 2003. XML for Data Architects: Designing for
Reuse and Integration, Morgan Kaufmann.
Holzner, S., 2001. Inside XSLT, Que.
RDF Primer. W3C Recommendation 10 February 2004,
OWL Web Ontology Language Guide. W3C Recommenda-
tion 10 February 2004, from
Batty, D. Controlled vocabulary and thesauri in support of
online information access. D-Lib Magazine, Novem-
ber, 1998
Clarke S.J. Search Engines for the World Wide Web: An
Evaluation of Recent Developments. Journal of Inter-
net Cataloging, sol.20, 3/4, 2000, 81-93.
Gilyarevsky R.S. UDC and its role in the development of
the information retrieval languages. Paper
synopses. Herceg Novi (Jugosl.), 1971. P. 6.
WEBIST 2007 - International Conference on Web Information Systems and Technologies