=Paper=
{{Paper
|id=Vol-2523/paper31
|storemode=property
|title=
Data Curation Approach to Management of Research Data. Use Cases for a Upgrade of the Thermophysical Database THERMAL
|pdfUrl=https://ceur-ws.org/Vol-2523/paper31.pdf
|volume=Vol-2523
|authors=Andrey Kosinov,Adilbek Erkimbaev,Georgy Kobzev,Vladimir Zitserman
|dblpUrl=https://dblp.org/rec/conf/rcdl/KosinovEKZ19
}}
==
Data Curation Approach to Management of Research Data. Use Cases for a Upgrade of the Thermophysical Database THERMAL
==
https://ceur-ws.org/Vol-2523/paper31.pdf
Data Curation Approach to Management of Research
Data. Use Cases for a Upgrade of the Thermophysical
Database THERMAL
Andrey Kosinov1, Adilbek Erkimbaev1, Geirgy Kobzev1, and Vladimir
Zitserman1
1 Joint Institute for High Temperatures, Russian Academy of Sciences, Russia
vz1941@mail.ru
Abstract. Procedures are considered to support extensive archives of digital da-
ta called “data storage”. Particular attention is paid to the support of scientific
data. It is shown that the activities aimed at updating the thermophysical
THERMAL database correspond to the approaches provided by the “data cura-
tion”. A communication system for metadata with external ontology is pro-
posed. The new version of metadata provides the possibility of multilateral as-
sessment of the origin, quality and status of scientific data. It is shown that the
use of new metadata provides a significant increase in the value of these stud-
ies.
Keywords: Research Data, Data Curation, Data Quality, Thermophysical Da-
tabase
1 Introduction
Digital data uploaded to repositories or databases require permanent procedures that
guarantee safety, quality top-level and enduring access to data. The set of such proce-
dures is the subject of particular activity of managing the use of data called data cu-
ration. This term is rarely used in the Russian literature, although all the necessary
actions for data integrity and management are of course performed to support digital
repositories or databases. The meaning and content of the concept of curation of data
can be revealed by referring to the history of its appearance. This concept originates
from the Museum's practice, which is traditionally based on the curator's work on
preservation, renovation and description of exhibits.
As for the term “data curation”, apparently, it first appeared in the article by Diana
Zorich [1], which pointed to the common problems facing libraries, museums and
research centers involved in supporting digital collections. According to [1], digital
archives, as well as their supporting tools (vocabularies, thesauruses, metadata)
should be regularly monitored and updated for data consistency, maintaining its quali-
ty, availability, etc., and activities in this direction is the essence of curating process.
Oddly enough, digital data is subject to erosion, as are physical artifacts, manu-
scripts or museum exhibits. It can be related to the use of outdated metadata, termi-
nology, dictionaries, formats, software, as well as the absence of references to more
Copyright © 2019 for this paper by its authors. Use permitted under Creative
Commons License Attribution 4.0 International (CC BY 4.0).
325
relevant documents or external resources. By analogy with engineering, it can be
considered as technological obsolescence (or deterioration) of the data structure, file
formats, software, etc. An unrecoverable failure of storage media (bit rot in IT slang)
during data storage is possible in parallel with content obsolescence.
History of the “Digital Curation” concept and its gradual adoption in data manag-
er’s community is considered by [2]. In particular, the concept of “data curation” here
is clearly separated from the narrower and service-oriented concepts: “archiving” and
“preserving”. Unlike the latter, curation involves not only the preservation and
maintenance of digital storage, but its indispensable enrichment by expanding the
functions and content. For example, a common and effective way of enriching content
is to place it in a wider context by linking the data set to thematically related re-
sources, so called Contextualizing.
Among the main objectives of data curation, as a rule, such are mentioned as their
storage, description, safety measures, the so-called cleaning, that is, monitoring and
restoring quality, as well as a number of other measures. The expanded definition of
the Digital Curation Center [3] covers all activities related to data management, start-
ing with data creation, digitization, documentation, and accessibility and future reuse.
The detailing of these processes, carried out in [4], allowed us to identify about 50
curation practices, most of which also fall into such categories as data preservation,
data cleaning, and finally, description in terms of the complex structure of metadata.
1 Curation of Research Data
The purpose of this report is to discuss the specific recipes foreseen in the framework
of digital curation in the implementation of the project for updating the Thermophysi-
cal Database THERMAL. The project considered in the report at the previous confer-
ence [5] includes: a significant expansion of the database volume due to the rejection
of the restrictions adopted at the stage of creation in the 70s of the last century; crea-
tion of tools for flexible variation of the data structure, reflecting the uniqueness of
objects and their characteristics; transition to a new platform that allows you to store
and process data of various structures and formats. A significant part of the activities
performed during the project, in fact, refers specifically to the data curation, as it
comes down to checking and correcting old documents in accordance with the newly
adopted format, vocabulary and requirements for data completeness and quality.
In general, curation refers to digital objects of arbitrary origin and kind. Therefore,
numerous measures for data preservation such as regular back up, defect detection at
the bit level, overcoming technological obsolescence of hardware or file formats are
applicable in all cases regardless of the content. In solving scientific problems, the
curation process provides not just conservation, but confirmation and reliable expan-
sion of previous data of the experiment or simulation. On the contrary, the absence or
poor quality of the curation process inevitably leads to the loss, distortion or misinter-
pretation of data.
A brief list of features and capabilities of “data curation” as applied to e-Science
(or Data-intensive science) was given by the Digital Curation Centre [6]. In this list,
the specificity of scientific data, to a certain extent, is taken into account at all stages
326
of the curation process. For example, long-term data storage may require replacement
of obsolescence of storage devices, which has already been encountered in astronomy.
At the stage of data cleaning (that is, data correction and updating), it is important to
establish links between different versions of evolving datasets or between primary and
secondary data. However, the most noticeable specificity of data curation is manifest-
ed in their description, that is, in the composition and structure of metadata.
2 Metadata (Update and Extension)
In general terms, metadata document the context and record information about how
the data was obtained and what processing and verification procedures were per-
formed during the retention period. There is an extensive literature on scientific
metadata and their use in various disciplines [7–9]. Metadata, accompanying subject
information, allow you to: identify a dataset with its position in the repository; define
access rules; describe the logical structure and data formats; to ensure the operation of
various data analysis tools. Metadata standards for different disciplines and types of
documents are collected in the catalog (rd-alliance.github.io/metadata-directory/) and
the “Disciplinary metadata” section of the Digital Curation Center
(www.dcc.ac.uk/resources/metadata-standards). Both mentioned sources contain also
references to domain-agnostic standards for formal description of digital resources
(e.g. the Dublin Core metadata set), or for the identification and citation of digital
resources (e.g. DataCite Metadata Store). Metadata for thermophysical properties
(ThermoML), characteristics of ordinary materials (MatML) and nanomaterials are
described in detail in [7, 10–12].
Regardless of the subject area, scientific metadata must satisfy a number of re-
quirements that guarantee sufficient completeness and accuracy in the presentation of
each data set. Relevant elements should provide: unambiguous identification of the
object of study; presentation of information about the source and data acquisition
method (research method, equipment, program code etc); uncertainty and data quality
information; linking with controlled vocabularies or ontologies; the possibility of
flexible adjustment to the features of the object and its characteristics. The expansion
of metadata carried out in the updating the database THERMAL, provides for the
implementation of each of these requirements, see Table 1.
First of all, in the progress of curation, the possibilities of identifying objects are
expanded, which can now include, along with inorganic substances, complex organ-
ics, natural and industrial materials, and so on. The “Identification’’ metadata ele-
ment provides, along with a stoichiometric formula, the use of several common names
(synonyms), as well as links to publicly available databases. The pointer to the data-
base and the corresponding identifier uniquely identify the object, providing, in addi-
tion, access to information that complements the information stored in THERMAL.
As an example in Fig. 1 the identification of the compound called epoxyethane
(oxirane, ethylene oxide) in the old and new versions of the database is shown.
327
Table 1. A comparison between old and updated metadata versions
Old metadata set New metadata structure
Unique record ID
Data type [bibl, full-text, factual]
Unique record ID Data status
[experiment or simulation, predicted, critical evaluated,
recommended, stale]
Research type [experimental, theory, simulation, review]
Source
Source Provenance [bibl, database, external agency]
Data origin
[method, equipment, software]
Abstract
Stoichiometric Identification
[common names, stoichiometric formula, public database
formula
ID]
Substance class Linking to ontology classes
[Linking to sub- classes of the Chemical_entity]
Properties
Linking to ontology classes
Properties type [Linking to sub- classes of the Quality]
Phase Linking to ontology classes
[Linking to sub-classes of the State_of_matter]
Linking to ontology classes
Phase transition
[Linking to sub-classes of the Transition]
Uncertainty [type, value]
Data Quality Data quality attributes (timeliness,
reliability, currency, completeness etc)
Data Features
[SubstanceFeatures, Sample, Influence Factors]
Full-text
Tables or equations
References
External documents
(from Web or Server)
328
Fig. 1. Identification of the substance “ethylene oxide” in the old (top) and new (bottom) ver-
sions of the database.
Another example is the natural mineral mullite, where variations of the elemental
composition allow the use of several stoichiometric formulas (for instance,
Al6Si2O13 и Al4SiO8), and the exact identification is provided by the record (URL:
www.webmineral.com/data/Mullite.shtml#.WaMcG_hJaUk) in the mineralogical
database WEBMINERAL.
A more complete identification is provided by linking metadata with ontology
classes, which includes entities that reflect the types of objects, their states and prop-
erties, Fig. 2.
In particular, chemical_entity identifies types of substances or materials, focusing
on the systematics adopted in chemistry (elements, oxides, acids ...), as well as on
categories determined by properties or by application (polymer, solution, mixture,
refrigerant, fuel ...). A pointer to subclasses in relation to the class mixture allows to
identify binary and multicomponent alloys and solutions, for example, such relevant
objects in thermophysics as air, humid air, combustion products, etc. The
State_of_matter class allows you to detail the phase and type of the crystal lattice
based on an extensive hierarchy of child classes.
Similarly, linking to classes that inherit the Quality and Transition classes reflects
the rich variety of physical properties inherent in an object and the phase transitions
that occur in it. It is essential that linking of metadata to ontology during the curation
provides unambiguous interpretation of terms and concept, and through editing of
ontology, the possibility of flexible adjustment in connection with the emergence of
new objects and concepts. For example, concepts such as second critical point [13],
329
topological insulator [14] or previously unknown allotropic forms of carbon (metal-
lic carbon, T-carbon) have recently been included.
Fig. 2. Fragment of the ontology – top level classes
3 Data Quality and Features
New curatorial perspectives have emerged with concepts such as “Data Quality” and
“Data Features” in the metadata set, see Table 1. Alternative choices for evaluation
the scientific data quality are discussed in detail in article [12]. It was shown that the
best way to assess the research data quality was to combine the traditional uncertainty
assessment with a certification procedure based on several quality attributes, every of
them represents some aspect of quality. It is necessary to select a specific metric for
the multidimensional certification of the dataset and the data compliance indicator for
each of the quality attributes. In many cases, it is useful to use a domain-agnostic
metrics reflecting quality factors that are important to data consumers, for example
accuracy, timeliness, reliability, completeness, relevancy, interpretability etc. Such an
approach to quality certification is most justified in interdisciplinary projects, for
example, when integrating thermal data with the performance characteristics of struc-
tural materials.
However, when you update the thermophysical database THERMAL, you can re-
duce the number of quality attributes. Data certification proposed by the authors [15]
in relation to physico-chemical properties is based on the use of three attributes: com-
pleteness of information about the state and preparing the sample in the experiment;
completeness of the method and measuring instruments description or codes and data
processing in the case of simulation; the consistency of the numerical data with
ground rules and regularities аs well as with previous fairly reliable measurements.
Combined with uncertainty assessment, certification identifies three main aspects of
330
data quality: accuracy, completeness, and consistency. The technique provides a gen-
eralized evaluation of the data set, by assigning each of the attributes of the quality
level (high, medium and low), focusing on compliance with the requirements of com-
pleteness and consistency. The expert gets the opportunity, based on this data curation
technique [15] select the data set of top-quality by assigning them the special Rec-
ommended status (for what to use the Data Stutus element). The information needed
to assess the data quality allows it to be carried out only for two types of data (Full-
text and Factual) specified in the Data Type element, Table 1. At the same time, for
unstructured data in the form of the text of an article (full-text data) it is justified to
conduct only certification with indication of quality attributes, but without its own
assessment of uncertainty.
Table 2. Examples of the non-standard data sets
Data Set Title Data Feature
SubstanceFeatures [amorphous,
Amorphous polymeric nitrogen-
toward an equation of state polymeric]
Melting point of high-purity ger- SubstanceFeatures [stable isotopes]
manium stable isotopes
Relationship between changes in
Sample [pellet]
the crystal lattice strain and thermal
conductivity of high burnup UO2 Influence Factors [high burnup]
pellets
Study of near-critical states of liq-
uid - vapor phase transition of metals Sample [porous]
by isentropic expansion method of
shock-compressed porous sample
Thermophysical properties of liq-
uid Co measured by electromagnetic Influence factors [field]
levitation technique in a static mag-
netic field
Phase diagram of water under an Influence factors [field]
applied electric field
Shock compression of preheated Influence factors [prehistory]
molybdenum
The concept of Data Features in a metadata set is based on other data evaluation
criterion. It allows you to select those data sets where there is any deviation from the
standard (i.e. an anomaly) in the characterization of the object or its properties. The
well-defined specificity (features) that distinguishes one dataset from another allows
you to overcome the inevitable contradiction between structured data and poorly for-
malized information hidden in context. As can be seen from the Table 1, “Data Fea-
tures’’ element includes three groups of the features: SubstanceFeatures, Sample,
331
Influence Factor. The first allows you to extend the traditional substance identifica-
tion, indicating the isomeric form, nonstoichiometry, isotopic composition, etc. The
indication Sample includes pointers to the features of the sample: shape, size, surface
condition, prehistory, etc. Finally, the Influence factor sign includes pointers to ex-
ternal factors that determine the experiment and properties of the substance: external
field, mechanical load, environment, radiation, etc. Some examples of non-standard
data sets from Table 2 illustrate the signs defining the specifics of a substance and its
properties. In so doing the specificity can be attributed to any data set of the three
types indicated in the Data Type element (Table 1), in contrast to the quality assess-
ment, depending on the type of data.
4 Data Cleaning and Preservation
The update procedure of the THERMAL database, includes (along with the volume
expansion) revision of old records based on the new metadata system. As a result, the
data curator needs to check the completed conversion. This activity, called “Data
Cleaning” (or cleansing), involves the detection and correction of “dirty”, that is
distorted or incomplete data [16]. Data pollution occurs for various reasons, among
which may be distortions of old records (input errors and duplication, incorrect data
distribution by fields) and errors when using new metadata to determine the object
and properties, see Table 1.
Previous data were entered without the use of controlled dictionaries, so the most
important task of cleaning is to eliminate ambiguity in terms and concepts, subjecting
them to ontology classes. For example, earlier a whole set of lexical elements (includ-
ing English and Russian terms) was associated with the concept of dynamic com-
pressibility, namely: Hugoniot, Hugoniot data, Hugoniot adiabat, shock Hugoniot,
shock adiabat, shock compression, release isentrope etc. Linking this whole set of
terms with a single ontology class (Dynamic Compressibility) eliminates synonymy
and dramatically facilitates semantic search. The same procedure requires that the
names of substances used in different records now appear in each record as a set of
common names, which eliminates search losses.
In addition to linking and unifying terms, data cleaning also provides for the cor-
rection of content, first of all, the correction (or fixation) of obviously obsolete nu-
merical data. At the same time, old records with correct filling of fields have histori-
cal value, even with outdated data. Therefore, measures are proposed to clean the
data, excluding the physical removal of the record. One of them is to add to the old
record an indication of the unreliability of the data by linking it with later (or network
resources), including reliable data. Another measure is to assign the obsolete data to
the sign “low quality” using the attribute “consistency” (see above “Data Quality”).
Finally, you can assign the status to “Stale” or “Recommended”, which allows you
to immediately separate high-quality data from clearly obsolete data during the
search.
Regular data cleaning inevitably requires entering them into a new context, ex-
plaining concepts, offering an introduction to the available handbooks, databases,
manuals, and so on. In the list of data curation practices such activity was named as
332
Contextualizing, i.e. “Use metadata to link the data set to related publications, disser-
tations, and/or projects that provide added context for how the data were generated
and why”. A new set of metadata (Table 1) allows linking with external resources
through two elements, “Identification” and “References”. The first uses a link to
Public Databases to accurately identify a substance with access to additional data.
For example, by including the reference CSID: 6114 (ID from Database ChemSpi-
der) in the “Identification” field, we can select “Ethylene oxide” from the group of
substances with the same formula C2H4O, gaining access to the reference data,
Fig. 3.
Fig. 3. Typical entry for ethylene oxide from ChemSpider (CSID 6114)
The References element (Table 1) provides a link to thematically pertaining re-
sources, but without requiring exact identification of the object. An example is the
linking with the Sacada database [Samara Carbon Allotrope Database,
http://sacada.sctms.ru/], which expands the set of information on carbon allotropes
presented in the THERMAL database.
Obviously, contextualization as an data curation practice requires the participation
of human experts, but not programmers. Therefore, the term “clearing” within the
333
framework of data curation means not only the rejection of dirty data, but also data
analysis and decision making. Thus, contextualization as an element of data curation
completely corresponds to the expression “added value to digital research data
throughout its lifecycle”, in response to the question “What is digital curation?”
4.1 Preservation
Long-term storage, to a large extent, a purely technological problem related to service
life of memory devices (a maximum of 100 years), the solution of which requires
significant funding. The required measures include regular backup and failure detec-
tion at the bit level. Particular attention should be paid to the physical protection of
the data storage, as the frequency of bit rot in data significantly increases due to pollu-
tion, thermal and radiation exposure and other external agents. Some of the protec-
tions activities are most adequate for research data. Among them are format migration
(i.e. consistent change in line with technological changes) and emulation recreating
outdated hardware and software on a modern platform.
When transferring the THERMAL database to the Big Data platform, the obsolete
ISO-2709 format, adopted in the 60s of the last century [17] as ISO standard for bib-
liographic description, is discarded. In the new version, documents in ISO format are
converted to structured text in JSON format, one of the most convenient for exchang-
ing data and metadata [5]. The advantage of a text document is the possibility of sim-
ple reading and editing, accessibility for human perception, convenient form of stor-
age and exchange of arbitrary structured information. The JSON-format (unlike ISO)
is convenient for storing factual information in the form of tables and nested struc-
tures, as well as numerous links to files of different formats (images, presentation
files, Web-pages, etc.), which is especially important when expanding the functions of
the THERMAL database. It is also important that the JSON format is a working ob-
ject for some platforms, in particular for Apache Spark, allowing for the exchange,
storage and queries for distributed data. There is already an experience of using struc-
tured text as a means of thermal properties data interchange [12].
Along with the obsolescence of formats, outdated software that is incompatible
with more modern platforms affects long-term storage. This fully applies to the data-
base THERMAL, built on the basis of the documentary Database Management Sys-
tem (DBMS) CDS/ISIS [17] with a fairly limited scope, mainly for the storage of
catalogue cards. The transition to the Apache Spark platform
(http://spark.apache.org/docs/) in combination with ontology-based data management
opens up much greater opportunities in storing and integrating data of arbitrary for-
mat, converted into JSON-format. In turn, ontology supports a single vocabulary of
all concepts, expanded by logical connections and axioms. An ontology encoded as an
OWL file becomes a control superstructure capable of semantic integration of hetero-
geneous data.
The arsenal of tools proposed in the project (JSON, Apache Spark, SPARQL)
meets modern standards for storing and processing heterogeneous data and is capable
of supporting interaction with many types of storage. Thereby, long-term data storage
334
will be provided with the possibility of painless migration to subsequent versions of
the software.
References
1. Zorich, D.M.: Data management: Managing electronic information: Data curation in mu-
seums. Museum Management and Curatorship 14 (4), 431 (1995).
2. Beagrie, N.: Digital curation for science, digital libraries, and individuals. The Internation-
al Journal of Digital Curation 1 (1), 3–16 (2006).
3. Abbott, D.: "What is Digital Curation?". DCC Briefing Papers: Introduction to Curation.
Edinburgh: Digital Curation Centre. Handle: 1842/3362 (2008). Available online:
http://www.dcc.ac.uk/resources/briefing-papers/introduction-curation
4. Johnson, L.R., et al.: How important are data curation activities to researchers? Gaps and
opportunities for academic libraries. Journal of Librarianship and Scholarly Communica-
tion, 6(General Issue), eP2198 (2018). Available online: https://doi.org/10.7710/2162-
3309.2198
5. Kosinov, A.V., Erkimbaev, A.O., Zitserman, V.Yu., Kobzev G.A.: Ontology-based meth-
ods of thermophysical data integration. In: XV Russian Conference (with international par-
ticipation) on Thermophysical Properties of Substances (RCTP-15), 103–104. Book of
Abstracts. Moscow, Russia, (2018).
6. Pennock, M.: Curating e-Science Data. DCC Briefing Papers: Introduction to Curation.
Edinburgh: Digital Curation Centre. Handle: 1842/3330 (2006). Available online:
http://www.dcc.ac.uk/resources/briefing-papers/introduction-curation
7. Yerkimbaev, A.O., Zitserman, V.Yu., and Kobzev, G.A.: The role of metadata in the crea-
tion and application of information resources on the properties of substances and materials.
Sci. Tech. Information Process 35 (6), 47–255 (2008).
8. Davenhall, C.: “Scientific Metadata”, DCC Digital Curation Manual, J. Davidson, S. Ross,
M. Day (eds), (2011). Available online: http://www.dcc.ac.uk/resources/curation-
reference-manual/scientific-metadata
9. Willis, C., Greenberg, J., and White, H.: Analysis and synthesis of metadata goals for sci-
entific data. J. American Soc. for Information Science and Technology 63 (8), 1505–1520
(2012).
10. Erkimbaev, A.O., Zitserman, V.Yu., Kobzev, G.A., and Fokin L.R.: The logical structure
of physicochemical data: problems of numerical data standardization and exchange. Rus-
sian Journal of Physical Chemistry A. 82 (1), 15–25 (2008).
11. Erkimbaev, A.O., Zitserman, V.Yu., Kobzev, G.A., and Trakhtenhers, M.S.: A universal
metadata system for the characterization of nanomaterials. Sci. Tech. Inf. Process 42 (4),
211–222 (2015).
12. Erkimbaev, A.O., Zitserman, V.Yu., and Kobzev, G.A.: The intensive use of digital data in
modern natural science. Automatic Documentation and Mathematical Linguistics 51 (5),
201–213 (2017).
13. Water Structure and Science. P7. Supercooled water has two phases and a second critical
point. Available online: http://www1.lsbu.ac.uk/water/phase_anomalies.html
14. Kane, C. and Moore, J.: Topological insulators. Physics World 32–36 (2011).
15. Eletskii, A.V., Erkimbaev, A.O., Kobzev, G.A., Trachtengerts, M.S., and Zitserman V.Y.:
Properties of nanostructures: data acquisition, categorization, and evaluation. Data Sci-
ence Journal 11, 126–139 (2012). Available online: https://www.jstage.jst.go.jp/browse/dsj
16. Rahm, E. and Do, H.H.: Data cleaning: problems and current approaches. IEEE Data Eng.
Bull. 23 (4), 3–13 (2000).
17. CDS/ISIS for Windows: Reference Manual (Version 1.31). Paris: UNESCO (1998).
335