The
construction of "correct classifiers has always played and continues to
play an important role in the automation of enterprises, which was demonstrated
at the 2023 conference “The Role of Standardization in the Digitalization of
Industry” [1]. For a large enterprise, the most important classification
objects are Resources, Works/services and Assets.
When
constructing an ontological model of an enterprise, classification plays an
important role for constructing first a taxonomy and then an ontology of the
enterprise as a whole, which was demonstrated at international ontology
conferences [2].
However,
correct classification plays no less a role in solving practical problems of
automation of specific organizations, especially large ones. When implementing
an enterprise master data, it is important to build a convenient and effective
tool for working with a large set of items. At the same time, an important part
of the work is to coordinate decisions made with stakeholders - business users,
who are sometimes far from information technology.
According to TOGAF recommendations [3] one of the tasks of an information system architect is to offer the most convenient views for perception by stakeholders with a wide range of interests. In order not to solve such a problem, it is necessary to offer a method for preparing such a viewpoint for each typical case.
A review of software tools for displaying geometric objects is given in an article by MEPhI scientists [4]. An overview of information visualization tools is given in the article [5] and other works by Bauman Moscow State Technical University employees. A representative collection of data visualization templates is presented by the Ferdio agency as a result of the data visualization project [6]. The article [7] provides a general set of template applications, including for displaying hierarchical classification.
This
paper describes the possibilities and results of practical application of
visualization tools for constructing a qualitative classification for a large
enterprise.
The
basic rules of hierarchical classification are given in the monograph [8]. Modern classification experts use approximately the same set of principles on
projects. An option taking into account the capabilities of modern systems is
given in [9] and also includes requirements of completeness, consistency, and
the presence of a classification feature at each level.
However,
in practice, when constructing real classifiers, problems arise that require
non-obvious solutions. Here are some of them:
- selection of
classification objects;
- selection of
classification features;
- ensuring the convenience
of the classifier while observing formal rules;
- implementation of
conflicting requirements for the composition of classes.
Let's
consider options for solving these problems using visualization tools using the
example of constructing a classifier of works and services.
To
correctly select classification objects, it is necessary to proceed from the
goals of constructing and further using the classifier and configuration of
integrated information systems. For example, detailing of services is usually
required when concluding contracts with suppliers or consumers of services,
when preparing specifications for equipment repair, less often for capital
construction, when accounting for costs. If there are a small number of areas,
this can be visualized using a Venn diagram (Figure 1).
Figure 1.
Initial record sets for the corporate directory “Works and services”
If
the task is to integrate three systems of the corresponding purpose, then the
completeness of the directory can be determined by the areas of intersection.
If several systems operate in each area, then a classifier covering the
combination of areas will be required. This solution will allow not only to
perform a consolidated analysis of works and services, but also to provide
uniform corporate accounting features for the entire enterprise.
To
select mandatory classification features, the requirements of the areas of use
may be considered. However, there is a risk of including specific attributes
inherent only to a given area. For example, in accounting, the activities of
contractors can be written off immediately in the reporting period as services,
or accumulated on cost objects as work performed. In general, such activities
can be both work and services, and such a sign cannot be used.
At the same time, if it is required to generate reports
according the content of services and such a sign can be used for
classification.
Another
difficulty is that feature must be made at each classification level. Here it
is useful to look at other directories of the organization for which the
general task of classifying all activities has already been solved. Typically,
the following classification criteria are used at the top level: divisions,
business processes, functional areas, main services and products.
Further, the features will depend on the composition of the classified records. The Section Layer Diagram (Figure 2) or by classification [6] «Icicle» allows you to depict not only the classification branches, but also the values of the features used diagram. The difference is that for our purposes it is expanded by 90 degrees and supplemented with important class - attributes and values of attributes of higher classes.
Figure 2.
Using a section layer diagram to show features and their meanings
One
of the common mistakes when building classifiers is the desire to “even out”
the number of levels and the length of classification branches at all costs. In
this case, the criterion of “convenience” is approximately the same complexity
of each level.
At
first glance, such a classifier is more convenient to use, since it looks more
streamlined visually. However, this is a deceptive impression. With this
approach, the principle of identifying a clear classification feature at each
level is violated. Complex subject areas are usually not symmetrical. In terms
of functional modeling, such an observation was described
in the methodology
for functional modeling [10], but this rule is also applicable to data
classification. Moreover, if you see a “beautiful” symmetrical classification,
then it is already visually clear that it is most likely incorrect.
On
the other hand, if we strictly follow the rule of one class - one attribute,
then in practice we can get a logically correctly constructed, but absolutely
inconvenient to use classifier.
The
solution here is to divide the construction of a classification tree into two
stages: the construction of a logical model and the option of physical model
for the final implementation in the system, similar to the approach to data
modeling [11]. At the logical level, it is useful to clearly record all
possible features and strict division at each level. At the physical level, you
can carefully combine the most branched sections, taking into account the
knowledge of the previous level, transferring the abbreviated features into the
class name.
In
Figure 3, when using a standard tree, we see that we can shorten the third
classification branch by combining the attributes “Object of transportation”
and “Mode of transport”.
In
a large enterprise, there may be different requirements for the required
classes. For example, in a purchasing department (or holding organization), it
is important to take into account the transportation of goods by mode of
transport in order to choose the most effective methods. And for a sales
department/organization it may be important to divide cargo transportation
services by product type for easier cost calculation.
In
principle, such a problem is solved by the Cartesian product of the required classes.
This approach is acceptable within one organization or for small volumes.
However, within the holding, each organization may require its own
classification feature at the same generalization.
This
limitation also has a solution method within a single hierarchical classifier.
It consists of system classification features, where the feature is the sign of
the class of the next level. In our example, the feature will be
“Classification Feature", and the values will be “Object of
transportation” and “Mode of transport” (Figure 4).
Figure 3.
Reducing the length of classification branches in the Tree diagram
Figure 4.
Using the system class feature
It
should be noted that this method is only applicable if
1) classes below this level
are used only locally; all corporate attributes can be linked at levels.
2) the use of parallel
classifications is significantly less effective: it requires additional
development in related systems, complicates data models, etc.
A more promising approach is to go beyond the use of a single hierarchy. More complex data models can help to reflect the necessary entities and relationships can help to find the most effective solution options. Three solution methods will be shown below, for the presentation of, which data modeling techniques are used. Among the many possible notations [12], [13] the most convenient notation has proven to be the visualization of relationships using Martin's notation (another name is “Crow's Foot”), which more clearly reflects the cardinality of the related entities and compactly represents a set of attributes.
3.1) Multidimensional classification
If
the same objects must be classified using different classification criteria,
then perhaps the option of moving to facet classification is suitable. A
conceptual data model showing this solution method is shown in Figure 5.
Figure 5.
Conceptual data model of multidimensional classification of services
Multidimensional
classification opens up additional possibilities for visual data analysis.
Figures 5 and 6 show examples of diagrams using multidimensional classification
of services.
Figure 6.
Volume histogram
Figure 7.
Surface diagram
3.2)
If the set of additional features for each subclass is unique and contains a
large number of linear values, we can use another approach that involves
storing our own sets of features for each class (Figure 6). It should be noted
that this approach is most effective for classifying materials. For more
information on solving problems of automating the directory of materials, see [14].[15]
Figure 6.
Conceptual data model for using selected sets of feature values for a class
3.3) Using individual analytics
To
illustrate this method, a logical data model is more suitable (Figure 7). In
this case, it shows that to obtain additional analytics (type of cargo), the
analyst may use a classifier of goods already used when preparing transactional
transportation documents.
Figure 7. Logical data model for linking services and cargo classes
This paper proposes options for solving problems of classifying works and services, which were successfully tested on projects for developing normative and reference information in a large Russian company with the participation of former MEPhI students [16], [17]IBM Product Master 12.0 platform was used as a tool for classifying and managing reference data in general, which has flexible means of managing the structure of reference data.
Among the classification management tools, the following can be noted:
– Convenient
visualization tools, including for multidimensional classifications
– Possibility
to import standard or define custom classifications
– View
and edit records using different classifications
– Managing
mapping of classes from different hierarchies
Additional
visualization tools include the following:
– Display
of configured workflows for reconciling records by different categories of
users in the form of sequence diagrams;
– Display
of reference data processing flows when setting up additional data quality
check procedures;
– Connecting
external visualization tools, such as maps to show the location of a directory
object with specified coordinates.
The
constructed classifier of works and services has about 400 classes with specified
classification features, is used to classify more than 6000 initial records of
information systems. 6 subject areas were used as data sources (Purchases,
Payments, Costs, Repairs, Investments, Accounting). The directory of works and
services is centrally maintained for more than 10 information systems of the
holding.
1. Kukshev V.I., "Prospects for the Development of Industrial Product Classifiers in the EAEU," 12 12 2023. [Online]. Available: http://www.rgtr.ru/data/events/2023/SMART%2C%2012.12.2023/8.%20%D0%9A%D1%83%D0%BA%D1%88%D0%B5%D0%B2%20%D0%92.%D0%98..pdf.
2. H. Hedden, “Turning a Taxonomy into an Ontology,” 17 11 2021. [Online]. Available: https://www.hedden-information.com/wp-content/uploads/2021/11/Turning-a-Taxonomy-into-an-Ontology.pdf.
3. The Open Group, “TOGAF,” The Open Group, April 2022. [Online]. Available: https://pubs.opengroup.org/togaf-standard/index.html.
4. M.N.Strikhanov, N.N.Degtyarenko, V.V.Pilyugin, E.E. Malikova, N.A. Matveeva, V.D.Adzhtev, A.A.Pasko, “Experience of computer visualization of nanostructures at MEPhI,” Scientific Visualization, vol. 1, no. 1, pp. 1-18, 2009.
5. O.V. Peskova, “On information visualization,” Bulletin of Bauman Moscow State Technical University. Series “Instrument Engineering”, 2012.
6. Ferdio, “Data Viz Project,” Ferdio, [ In Internet ]. Available: https://datavizproject.com. [Accessed: 30 07 2024].
7. I.K. Romanova, "MODERN METHODS OF MULTIDIMENSIONAL DATA VISUALIZATION: ANALYSIS, CLASSIFICATION, IMPLEMENTATION, APPLICATIONS IN TECHNICAL SYSTEMS," Science and Education: scientific publication of Bauman Moscow State Technical University, No. 3, pp. 133-167, 2016.
8. Ayvazyan S.A., Bezhaeva Z.I., Staroverov O.V., Classification of multivariate observations, Moscow: Statistics, 1974.
9. "Requirements for a hierarchical classification system," [Online]. Available: https://studylib.ru/doc/4675234/trebovaniya-k-ierarhicheskoj-sisteme-klassifikacii. [Accessed: 30 07 2024].
10. K. M. David Mark, Methodology of structural analysis and design: Translated from English, Moscow: Original layout: "Meta-technology", Printing: GMP "First Model Printing House", 1993.
11. Mark Mosley, Michael Brackett, Susan Earley, Deborah Henderson, The DAMA Guide to the Data Management Body of Knowledge, NJ USA: DAMA International, 2010.
12. M. Nizhelskaya, N. Klassen, A. Pavlova, “Entities and Relationships: How and Why System Analysts Create ER Diagrams,” 17 05 2023. [Online]. Available: https://practicum.yandex.ru/blog/chto-takoe-er-diagramma/.
13. "Entity-Relationship Model Notations (ER Diagrams)," 11 09 2021. [Online]. Available: https://pro-prof.com/archives/8126.
14. Dzengelevsky A.E., Mogilat A.S., “Classifiers of the KSSS: now also finished products,” ITime – Information technologies in the fuel and energy complex, v. 2(12), pp. 26-29, 2010.
15. Dzengelevsky A.E., “Levels of Generalization in Accounting for Inventory,” Actual Problems of Humanities and Natural Sciences, Vol. 12 (59), No. I, pp. 82-90, 2013.
16. S. A. Dzengelevsky A. E., “History of the KSSS: Unified Corporate System of Dictionaries and Reference Books,” ITime – Information Technologies in the Fuel and Energy Complex, 2007.
17. H. Sh. Dzengelevsky A.E., “Experience in the creation and development of a corporate system for managing normative and reference information,” Scientific and Methodological Journal “Inter-Industry Information Service”, v. 2(12), pp. 22-26, 2012.
