Data Quality Evaluation of Scientific Datasets
A Case Study in a Policy Support Context
Antonella Zanzi
1,2
and Alberto Trombetta
1
1
Dipartimento di Scienze Teoriche e Applicate, Universit`a degli Studi dell’Insubria, via Mazzini 5, 21100 Varese, Italy
2
Joint Research Centre, European Commission, via Enrico Fermi 2749, 21027 Ispra, Italy
Keywords:
Data Quality, Data Constraints, Data Dependencies, Order Dependencies, Existence Constraints.
Abstract:
In this work we present the rule-based approach used to evaluate the quality of scientific datasets in a policy
support context. The used case study refers to real datasets in a context where low data quality limits the
accuracy of the analysis results and, consequently, the significance of the provided policy advice. The applied
solution consists in the identification of types of constraints that can be useful as data quality rules and in
the development of a software tool to evaluate a dataset on the basis of a set of rules expressed in the XML
markup language. As rule types we selected some types of data constraints and dependencies already proposed
in data quality works, but we experimented also the use of order dependencies and existence constraints. The
case study was used to develop and test the adopted solution, which is anyway generally applicable to other
contexts.
1 INTRODUCTION
Awareness of the critical importance of data quality
issues has grown rapidly in the last decades in scien-
tific, industrial and governmental contexts. For new
archives or data collections the prevention is obvi-
ously a fundamental aspect, still, issues can arise in
the data management process, when data are derived
from other data, and when a dataset results from the
integration of multiple data sources. Moreover, in
some cases, it is necessary to use datasets received
from external unverified sources without having any
control on the original acquisition process.
An approach is to attempt to cleanse data. Data
cleaning is not a simple task, it is highly context de-
pendent, and, in many cases, data can be cleaned ef-
fectively only with some human intervention, since
fully automated cleaning procedures could lead to a
loss of useful information.
The discovery of incorrect data, the first necessary
step in data cleaning, is – in most cases – challenging;
moreover,when the presence of errors is recognized it
is not always feasible to trace back the correct values
(e.g., detecting inconsistencies among data may not
be sufficient to determine which record is at fault) or
it is not possible to correct a dataset (e.g., changes in
the original dataset are not allowed).
Another relevant aspect related to the quality of a
dataset is its fitness for the intended purpose, which
is one of the definition provided for data quality, as
for example in (Juran, 1964) ”Data are of high qual-
ity if they are fit for their intended use in operations,
decision-making, and planning” and in (Shanks and
Corbitt, 1999), where the authors adopted the defini-
tion of quality as ”fitness for purpose”. It may occur
that a dataset containing correct data is not useful in a
specific context, for example because a different level
of detail is requested, or for insufficient data coverage.
In these cases, efforts need to be devoted in assessing
the quality level of datasets in order to evaluate their
fitness for the intended purpose.
In the present work, we focus on the evaluation of
the quality of scientific datasets
1
considering mainly
the accuracy, consistency and coverage data charac-
teristics and we use, as case study, some datasets col-
lected by the Joint Research Centre of the European
Commission, the Commission’s in-house science ser-
vice having the mission to provide European Union
(EU) policies with independent, scientific and tech-
nical support. The adopted solution consists in the
identification of types of constraints that can be use-
ful as data quality rules and in the development of a
software tool to evaluate a dataset on the basis of a
specified set of rules.
1
For a discussion about scientific dataset definitions re-
fer to (Renear et al., 2010).
167
Zanzi A. and Trombetta A..
Data Quality Evaluation of Scientific Datasets - A Case Study in a Policy Support Context.
DOI: 10.5220/0004476401670174
In Proceedings of the 2nd International Conference on Data Technologies and Applications (DATA-2013), pages 167-174
ISBN: 978-989-8565-67-9
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)
As constraints and their enforcement play a key
role in the maintenance of data integrity into a
database, rules and their verification can play a key
role in the evaluation and assessment of the consis-
tency of a dataset. The consistency of a dataset can
be defined in terms of constraints, and inconsisten-
cies in the dataset can be detected as violations of
these constraints. Classic integrity constraints are rel-
evant for data quality and for data cleaning, however,
they do not capture all the data quality concerns. For
this reason, new forms of quality constraints, which
can be considered an extension of usual semantic in-
tegrity constraints in databases, are proposed and in-
vestigated. Moreover, data quality rules can play a
further role: they can help in verifying the suitability
of a dataset to be used for a certain purpose.
As data quality rule types to be used in our tool,
first of all, we selected some types of data constraints
and dependencies already proposed in data quality
works, as for example association rules, Functional
Dependencies (FDs) and Conditional Functional De-
pendencies (CFDs), the latest being FDs holding on
a subset of the relation instance and recently intro-
duced in the data cleaning context (Bohannon et al.,
2007). In addition, we considered
order dependencies
(Ginsburg and Hull, 1983) and
existence constraints
(Atzeni and Morfuni, 1986) because they can be use-
ful in the data quality evaluation context, even if, to
our knowledge, they have not yet been used in the
data quality context.
2 THE CASE STUDY CONTEXT
In the European Union the fisheries sector is managed
through the Common Fisheries Policy (CFP). Since
the CFP was first established in 1983, scientific advice
has increasingly become a major part of the fisheries
management decision-making process.
The are several impediments to the rational con-
trol of marine resources, and one of them is inade-
quate data. Fisheries management decisions are often
based on population models, but the models need data
to be accurate (Chen, 2003).
In order to allow a pan-European set of data to be
used for policy advice, the Commission Regulation
(EC) No. 665/2008 (European Commission, 2008)
established the Data Collection Framework (DCF), a
Community framework for the collection of data in
the fisheries sector. The framework requires Mem-
ber States to collect biological and economic data of
many European fisheries and related fisheries sectors,
and to provide access to these data for fisheries man-
agement advice, scientific publication, public debate
and stakeholder participation in policy development.
The collected fisheries data can be divided in the fol-
lowing datasets:
• Economic Data: employment, expenditure, capi-
tal and investments, fishing rights, and direct sub-
sidies;
• Biological Data: length and age distribution, ma-
turity data by age and length, sex ratio by age and
length, discards, and discards length distribution;
• Effort Data: days at sea and energy consumption;
• Transversal Data: capacity, landings, and prices;
• Data from Surveys (i.e., sampling at sea).
Data are normally aggregatedat various levels, for ex-
ample by year, by area, by fleet segment, by mesh
size, and special management conditions. Details
about the datasets, which are not reported here for
lack of space, can be found in a dedicated Web
site (https://datacollection.jrc.ec.europa.eu)where the
collected data are published. The Joint Research Cen-
tre (JRC) on behalf of the Directorate-General for
Maritime Affairs and Fisheries of the European Com-
mission, collects, checks and maintains the fisheries
data reported by EU Member States in the framework
of the DCF. The data quality checks performed at JRC
is concretely helping Member States in assessing the
quality of the data they provide and also in improving
the quality of their data management process. De-
tailed information about the activities carried out by
JRC in the context of the DCF can be found in the
already mentioned Web site.
To illustrate the processes involved in the fisheries
data submission from Member States, with particu-
lar emphasis on the activities addressing data qual-
ity verification and improvement, we use an Infor-
mation Product Map (IP-MAP)
2
, which is a graphi-
cal model for the description of information produc-
tion processes (Shankaranarayan et al., 2000). The
purpose of the IP-MAP shown in figure 1 is to po-
sition the step of the evaluation of data quality rules
(highlighted in gray in the diagram) in the case study
data management cycle. During the data submission
procedure several preliminary checks are carried out,
while other checks are performed after the submis-
sion procedure is concluded and the received data are
stored in the staging database. The following steps re-
lated to data quality aspects are shown in the diagram:
• Data Domain Check - Each value is checked
against the assigned domain.
• Duplicate Detection - Duplicate records and
records referring to the same entities (e.g., the
2
The list of the constructs that can be used to build IP-
MAPs can be found in the appendix.
DATA2013-2ndInternationalConferenceonDataManagementTechnologiesandApplications
168
Figure 1: An IP-MAP for the case study.
same fleet segment) but with different values are
identified.
• Evaluation of Data Quality Rules - In the diagram,
the data quality block labeled evaluation of DQ
rules, refers to the step of data quality evaluation
based on a set of rules mainly provided by experts
in fisheries data.
• Coverage Check - For each country, coverage
checks are carried out to ensure that all the re-
quired data for each fleet segment has been sub-
mitted. To establish how many and which fleet
segments should be reported by each country, the
National Programmes of each Member State are
consulted.
• Comparison with other Data Sources - Data from
the EU Sea Fishing Fleet Register and from the
Eurostat’s archives are used to assess the consis-
tencies of some of the provided data aggregated at
national level.
3 DATA QUALITY RULES
AND THEIR EVALUATION
We addressed two main targets: the expression of
quality concerns and their evaluation against concrete
instances. We adopted the solution to specify the rules
using the XML markup language (the DTD for the
used XML structure is shown in the appendix) and
we developed a tool to evaluate a dataset against a
specified set of rules. The developed tool, which is
Java technology based, provides functionalities to ex-
press data quality rules and to evaluate them against a
dataset stored in a relational database, identifying the
data subsets that do not comply with the defined rules.
In order to facilitate the input of the rules, predefined
templates were built for each type of rule accepted
by the tool; after a template has been completed with
the details of a rule, the correct XML code for that
rule is automatically generated. To validate the XML
format, in addition to DTD and XML Schema, we
used Schematron (van der Vlist, 2007), which is a
structural based validation language. The data to be
checked are stored in a relational database; in partic-
ular, for the case study we used PostgreSQL. In the
rules, it is necessary to specify the tables containing
the data; instead of tables, views can be used as well,
allowing more flexibility in the definition of data con-
straints. For each defined rule, the tool identifies all
the records that do not comply with it and stores them
in views or tables inside the same database containing
the source dataset.
3.1 Types of Rules
The rules managed by the developed tool are classi-
fied in the following types:
Functional Dependency Rule. Functional depen-
dencies among attributes of the same relation can
be defined with this type of rule, including CFDs
(i.e., FDs holding on a subset of the relation
instance) and CFD
p
s, the latest being CFDs
specifying patterns of data values with predicates
(6=, <, >, ≤, ≥) (Chen et al., 2009). The
DataQualityEvaluationofScientificDatasets-ACaseStudyinaPolicySupportContext
169
left-hand side and the right-hand side of the rule
can contain one or more attributes. Moreover, it
is possible to add conditions connected by means
of boolean operators in order to express CFDs
and CFD
p
s (in the XML rule format, conditions
are recorded using the tag
when
).
Conditional Constraint Rule. This type of rule can
specify a constraint that has to be valid if a given
condition is satisfied and can be used to define any
constraint that can be written in the form
if-then
among attributes of the same relation. Associ-
ation rules, constant CFDs and constant CFD
p
s
(namely, CFDs and CFD
p
s with constant patterns
only) can be expressed with this kind of rule. The
if
and
then
components of the rule can contain
one or more conditions, connected by means of
boolean operators, on attributes of the same table.
Differential Dependency Rule. For this rule we re-
fer to a new kind of dependency called
differential
dependency
. It was proposed in (Song and Chen,
2011) and it can be used to specify constraints
on distances over attributes of the same relation.
In the developed tool, we limited the application
of differential dependencies only to numerical at-
tributes, but we introduced the possibility to ap-
ply these dependencieson non-overlappingsets of
tuples on the basis of specified attributes (which
in the XML format are expressed using the tag
partition on
).
Check Constraint Rule. A check constraint rule
consists in boolean conditions, which can contain
arithmetic operators, on single attributes or among
attributes belonging to the same or to different re-
lations. When more than one table is involved, it
is required to specify the criteria to be used to join
the tables themselves (in the XML rule format,
joining criteria are labeled with the tag
join on
).
Existence Constraint Rule. Referring to the defini-
tion of existence constraints introduced in works
related to databases with incomplete information
(Atzeni and Morfuni, 1986), we defined existence
constraint rules among attributes belonging to the
same or to different relations.
Order Dependency Rule. Adopting the order de-
pendency definition provided in literature (Gins-
burg and Hull, 1983), we extended it defining a
rule type to express direct and inverse order con-
straints among attributes belonging to the same re-
lation.
In the rest of the paragraph, we will give more de-
tails about the existence constraint and order depen-
dency rules managed by the tool, providing examples
of SQL statements that can be used to identify non-
complying data in a dataset. Note that in the shown
SQL statements, XML tags are referred using Xpath-
like expressions (for the complete structure refer to
the DTD shown in the appendix).
3.1.1 Existence Constraint Rules
Two types of existence constraints, respectively called
existence constraints and disjunctive existence con-
straints, have been defined in works related to
databases with incomplete information (Atzeni and
Morfuni, 1986). In contrast with the definitions pro-
posed in literature, the implemented rule manages
only attributes instead of sets of attributes in the left-
hand side of disjunctive existence constraints and in
both sides of existence constraints.
More formally, given a relation schema R(U) and
a relation instance r over R, this type of rule can ex-
press dependency-like and disjunctive-like existence
constraints defined as follows:
• A dependency-like existence constraint A → B
(read A requires B), where A, B ∈ U, holds over
r if, for each tuple t ∈ r, t[A] 6= null implies
t[B] 6= null.
• A disjunctive-like existence constraint A → S,
where A ∈ U and S = {Y
1
, . . . ,Y
n
} is a set of sets
of attributes (with Y
1
, . . . ,Y
n
⊆ U), holds over r if,
for each tuple t ∈ r, t[A] 6= null then ∃ Y
i
∈ S (with
1 ≤ i ≤ n) such that ∀ B ∈ Y
i
t[B] 6= null.
In addition, this kind of rule allows the definition of a
not null
constraint on a single attribute and a bidirec-
tional existence constraint between a pair of attributes
(if one of the two attributes exists in the dataset, the
second has to exist as well) contained in the same ta-
ble or in different tables.
The scope of the rule can be limited to a subset of
the relation instance through conditions (which can be
expressed in the XML format using the tag
when
). In
addition, when two tables are referred, it is required
to specify the criteria to be used to join them (in the
XML rule format, joining criteria are labeled with the
tag
join on
).
General SQL statements to retrieve non-
complying records for dependency-like, disjunctive-
like and bidirectional existence constraints, when
the involved attributes belong to different tables, are
listed in the following:
SELECT //rule_ec//column_name
FROM //lhs_ec//table_name
LEFT OUTER JOIN //rhs_ec//table_name
ON (//rule_ec/join_on)
WHERE //lhs_ec//column_name IS NOT NULL
AND //rhs_ec//column_name IS NULL
AND //rule_ec/when
DATA2013-2ndInternationalConferenceonDataManagementTechnologiesandApplications
170
SELECT //rule_ec//column_name
FROM //lhs_ec//table_name
LEFT OUTER JOIN //rhs_ec/table_name
ON (//rule_ec/join_on)
WHERE //lhs_ec//column_name IS NOT NULL
AND
(//rhs_ec/disj_attr[i]/column_name[1] IS NULL
OR
//rhs_ec/disj_attr[i]/column_name[k] IS NULL)
AND //rule_ec/when
SELECT //rule_ec//column_name
FROM //rule_ec/column[1]/table_name
FULL OUTER JOIN //rule_ec/column[2]/table_name
ON (//rule_ec/join_on)
WHERE (//rule_ec/column[1]/column_name IS NULL
AND
//rule_ec/column[2]/column_name IS NOT NULL)
OR (//rule_ec/column[2]/column_name IS NULL
AND
//rule_ec/column[1]/column_name IS NOT NULL)
AND //rule_ec/when
In the case study, this kind of rule was used to
express constraints like the following one: ”For every
active fleet segment, landing data (weight and value)
have to be provided”.
3.1.2 Order Dependency Rules
This kind of rule can be used to define order depen-
dencies among attributes: given a relation schema
R(U), an order dependencyX → Y (X,Y ⊆ U) holds if
an order over the values of each attribute in X implies
an order over the values of each attributes of Y.
With this rule type it is possible to express differ-
ent orderings for the attributes contained in the left-
hand side and the right-hand side of the dependency,
allowing the definition of both direct and inverse or-
der dependencies.
More formally, given a relation schema R(U) and
a relation instance r over R, this type of rule can be
used to define a dependency X → Y (X,Y ⊆ U) such
that:
• X →
Y denotes a
direct
order dependency if, for
every pair of tuples s andt ∈ r, s[X] t[X] implies
s[Y] t[Y], where s[X] t[X] if s[A] ≤ t[A] ∀ A ∈
X, and s[Y] t[Y] if s[B] ≤ t[B] ∀ B ∈ Y.
• X →
Y denotes an
inverse
order dependency if,
for every pair of tuples s and t ∈ r, s[X] t[X]
implies s[Y] t[Y], where s[X] t[X] if s[A] ≤
t[A] ∀ A ∈ X, and s[Y] t[Y] if s[B] ≥ t[B] ∀ B ∈ Y.
The scope of the rule can be limited to a subset of
the relation instance through conditions (which can
be expressed in the XML format using the tag
when
).
In addition, it is possible to apply the rule on non-
overlapping sets of tuples on the basis of specified at-
tributes (which in the XML format are expressed us-
ing the tag
partition on
).
The following SQL statement
3
identifies the
records that do not satisfy a
direct
order dependency
rule and provides, for each selected record, the num-
ber of failed comparisons.
WITH tmpView AS (
SELECT DISTINCT //lhs/column_name,
//rhs/column_name,
//rule_od/partition_on/column_name as partCol
FROM /rule_definition/table_name
WHERE //rule_od/when
)
SELECT tableName.//lhs/column_name,
tableName.//rhs/column_name, count(*)
FROM tmpView,
/rule_definition/table_name as tableName
WHERE //rule_od/partition_on/column_name[j] =
tmpView.partCol[j] AND //rule_od/when
AND ((tableName./lhs/column_name[i] <=
tmpView.//lhs/column_name[i]
AND (NOT tableName.//rhs/column_name[1] <=
tmpView.//rhs/column_name[1]
OR NOT tableName.//rhs/column_name[k] <=
tmpView.//rhs/column_name[k]))
OR (tmpView.//lhs/column_name[i] <=
tableName./lhs/column_name[i]
AND (NOT tmpView.//rhs/column_name[1] <=
tableName.//rhs/column_name[1]
OR NOT tmpView.//rhs/column_name[k] <=
tableName.//rhs/column_name[k])))
GROUP BY tableName.//lhs/column_name,
tableName.//rhs/column_name
An example of order dependency used in the case
study is the following one: ”In a survey at sea, an in-
verse order dependency holds among registered tem-
peratures and hauling depths”.
3.2 The used Metric
In the case study context, in order to quantify the qual-
ity of the data provided by Member States, the main
used metric is strictly connected with the adopted data
quality rules; in fact, the used metric is based on the
number of records satisfying the rules specified for
the dataset.
Given a set of rules, and called respectively:
– N the total number of the rules
– w
r
the weight (relevance) of the rule r
– S
r
the number of records satisfying the rule r
– U
r
the number of records not satisfying the rule r
I
DQ
w
=
N
∑
r=1
−
I
r
∗ w
r
N
I
r
=
U
r
S
r
+U
r
(1)
3
The SQL
WITH
clause implements the Common Table
Expression defined in the SQL:1999 standard.
DataQualityEvaluationofScientificDatasets-ACaseStudyinaPolicySupportContext
171
The values of I
DQ
w
will be 0 or a negative number,
where 0 is the best value for the index.
Due to the types of rules used in the present work,
this kind of measure mainly refers to the accuracy,
consistency and coverage data characteristics and, in
the case study context, it was used to compare differ-
ent datasets of different years, or subsets of the whole
dataset (e.g., data by fleet segment or Member State).
4 RELATED WORKS
Data quality issues have been faced from different
point of views and with several approaches.
Prevention approaches refer, for example, to
proper database design and to the use of integrity con-
straints, but also to appropriate business process man-
agement avoiding the generation of low data quality.
Common diagnostic approaches are database profil-
ing and exploratory data analysis (Dasu and John-
son, 2003). Currently used corrective approaches,
which are provided by the available data cleaning
tools, comprise data cleaning methods for attributes
(Hellerstein, 2008), duplicate detection techniques
(Elmagarmid et al., 2007), and virtual repair methods
(Bertossi and Chomicki, 2003). Some tools concen-
trate on a specific domain, such as cleaning names and
address data, or on a specific cleaning phase, such as
duplicate detection; at the contrary, in the Extraction
Transformation Load (ETL) tools, which aim at help-
ing in the construction of data warehouses, the pro-
vided built-in data cleaning capabilities cover a large
part of the data transformation needs. A survey with
a detailed feature comparison of data cleaning tools
both of academic and industrial origin can be found
in (Barateiro and Galhardas, 2005); while for an in-
troduction to the issues faced and approaches used in
ETL tools refer to (Vassiliadis, 2009).
Talking about
data quality rules
one can refer to
rules used with different purposes. Integrity con-
straints are examples of rules normally enforced on
databases and rules are used in data cleaning and ETL
tools in order to express transformations to be applied
to data.
A business rule management system (Bajec et al.,
2000) is a software system used to define, execute
and monitor the decision logic that is used by oper-
ational systems within an organization. The decision
logic is represented through business rules meant to
capture knowledge of constraints in an organization.
The main focus of this kind of systems is not on data
quality, however, they can contribute in improving
the quality of the data produced by the business pro-
cesses.
Some literature works proposed to use association
rules extracted from a dataset as a means to discover
dirty data; for example, in (Hipp et al., 2001) the au-
thors present a rule-based data mining approach for
outlier detection.
Recent works proposed to use FDs and exten-
sions to FDs for data cleaning purposes in the con-
text of relational databases. For example, in (Pivert
and Prade, 2010) the authors consider the case where
dirtiness corresponds to the violation of one or sev-
eral FDs. The use of the recently proposed CFDs as
a method for inconsistency detection and repairing is
discussed in (Cong et al., 2007). In
Semandaq
(Fan
et al., 2008), a tool using CFDs for data cleaning pur-
poses, users can specify CFDs through the drag and
drop functionality provided in the user interface. An-
other tool, called
Data Auditor
, is presented in (Golab
et al., 2010) and supports more types of constraints
(i.e., CFDs,
conditional inclusion dependencies
and
conditional sequential dependencies
) for testing data
inconsistency and completeness. CFD
p
s, which ex-
tend CFDs, were introduced, as already mentioned,
in (Chen et al., 2009); in the same work, the authors
describe an approach to automatically generate SQL
queries to select all the data that violate a set of CFD
p
s
encoded in relational tables.
To the best of our knowledge, existence con-
straints and order dependencies have not yet been
used in the context of data quality evaluation.
Existence constraints were introduced in works
related to databases with incomplete information
(Atzeni and Morfuni, 1986); moreover, in the
database design context, the term existence depen-
dency was used as a synonym of participation con-
straint (Elmasri and Navathe, 2000), a structural con-
straint indicating that the existence of an entity de-
pends on its being related to another entity.
Order dependencies, instead, were introduced for
the first time in the context of database systems by
(Ginsburg and Hull, 1983), later
pointwise ordered
functional dependencies
and
lexicographical ordered
functional dependencies
were formally defined in
(Ng, 1999). Moreover, in a recent work, order depen-
dencies were used in the query optimization context
(Szlichta et al., 2012).
Assessment of data quality can be performed in
relation to several data quality characteristics and, for
each of the selected characteristic, specific variables
to be measured can be chosen; examples of metrics
of this kind are shown in (Lee et al., 2006). The
metric used in our work, instead, is more compara-
ble with the types of metrics (also mentioned in (Lee
et al., 2006)) developed to measure data adherence to
integrity constraints in relational databases.
DATA2013-2ndInternationalConferenceonDataManagementTechnologiesandApplications
172
5 CONCLUSIONS
In order to deal with inconsistencies among data and
to evaluate the data quality level of a dataset, we pro-
posed an approach based on rules expressed with an
XML-based syntax and we developeda tool to deploy
such rules on a dataset. The proposed approach al-
lows a user to easily define different kinds of rules in
the same environment, without to deal with direct ma-
nipulation of XML trees. In fact, in order to express
data quality rules, a user needs to know the database
schema of the dataset and the types of rules managed
by the tool. The developed tool manages a predefined
number of rule types, but it can be easily extended in
order to deal with other types of rules.
As further work, we seek to perform an assess-
ment of the proposed approach on larger – and possi-
bly different – datasets, in order to validate it on other
applicative domains.
ACKNOWLEDGEMENTS
The authors acknowledge the contribution of the JRC
colleagues involved in the DCF activities, while be-
ing solely responsible for possible incomplete or er-
roneous statements.
Disclaimer. The content of this work reflects only
the opinion of the authors and may not be regarded as
stating an official position of the European Commis-
sion.
REFERENCES
Atzeni, P. and Morfuni, N. (1986). Functional dependencies
and constraints on null values in database relations.
Information and Control, 70(1):1–31.
Bajec, M., Krisper, M., and Rupnik, R. (2000). Us-
ing business rules technologies to bridge the gap be-
tween business and business applications. In Rechnu,
G., editor, Proceedings of the IFIP World Computer
Congress, pages 77–85.
Barateiro, J. and Galhardas, H. (2005). A survey of data
quality tools. Datenbank-Spektrum, 14/2005:15–21.
Bertossi, L. and Chomicki, J. (2003). Query answering in
inconsistent databases. In Chomicki, J., Saake, G.,
and van der Meyden, R., editors, Logics for Emerging
Applications of Databases, pages 43–83. Springer.
Bohannon, P., Fan, W., Geerts, F., Jia, X., and Kementsiet-
sidis, A. (2007). Conditional functional dependencies
for data cleaning. In Proceedings of the Int’l Confer-
ence on Data Engineering, pages 746–755.
Chen, W., Fan, W., and Ma, S. (2009). Analyses and
validation of conditional dependencies with built-in
predicates. In Bhowmick, S., Kung, J., and Wag-
ner, R., editors, Proceedings of the Int’l Conference
on Database and Expert Systems Applications, vol-
ume 5690 of LNCS, pages 576–591. Springer-Verlag.
Chen, Y. (2003). Quality of fisheries data and uncertainty in
Stock Assessment. Scientia Marina, 67(Suppl. 1):75–
87.
Cong, G., Fan, W., Geerts, F., Jia, X., and Ma, S. (2007).
Improving data quality: Consistency and accuracy.
In Proceedings of the Int’l Conference on Very Large
Data Bases, pages 315–326. VLDB Endowment.
Dasu, T. and Johnson, T. (2003). Exploratory Data Mining
and Data Cleaning. Wiley.
Elmagarmid, A., Ipeirotis, P., and Verykios, V. (2007). Du-
plicate record detection: A survey. IEEE Transactions
in Knowledge and Data Engineering, 19(1):1–16.
Elmasri, R. and Navathe, S. (2000). Foundamentals of
Database Systems (3rd edition). Addison-Wesley.
European Commission (15 July 2008). Commission Reg-
ulation (EC) No. 665/2008 of 14 July 2008. Official
Journal of the European Union.
Fan, W., Geerts, F., and Jia, X. (2008). Semandaq: A data
quality system based on conditional functional depen-
dencies. In Proceedings of the Int’l Conference on
Very Large Data Bases, pages 1460–1463. VLDB En-
dowment.
Ginsburg, S. and Hull, R. (1983). Order dependency in
the relational model. Theoretical Computer Science,
26:149–195.
Golab, L., Karloff, H., Korn, F., and D., S. (2010). Data
Auditor: Exploring data quality and semantics using
pattern tableaux. Proceedings of the VLDB Endow-
ment, 3(2):1641–1644.
Hellerstein, J. (2008). Quantitative data cleaning for large
databases. Report for the United Nations Economic
Commission for Europe (UNECE), 42 pp.
Hipp, J., G¨untzer, U., and Grimmer, U. (2001). Data quality
mining - making a virtue of necessity. In Proceedings
of the ACM SIGMOD Workshop on Research Issues in
Data Mining and Knowledge Discovery, pages 52–57.
Juran, J. (1964). Managerial breakthrough. McGraw-Hill,
New York.
Lee, Y., Pipino, L., Funk, J., and Wang, R. (2006). Journey
to Data Quality. The MIT Press.
Ng, W. (1999). Order functional dependencies in relational
databases. Information Systems, 24(7):535–554.
Pivert, O. and Prade, H. (2010). Handling dirty databases:
From user warning to data cleaning - towards an in-
teractive approach. In Deshpande, A. and Hunter, A.,
editors, Proceedings of the Int’l Conference on Scal-
able Uncertainty Management, volume 6379 of LNAI,
pages 292–305. Springer-Verlag.
Renear, A., Sacchi, S., and Wickett, K. (2010). Definitions
of dataset in the scientific and technical literature. In
Grove, A., editor, Proceedings of the American So-
ciety for Information Science and Technology Annual
Meeting, volume 47(1), pages 1–4.
Shankaranarayan, G., Wang, R. Y., and Ziad, M. (2000).
Modeling the manufacture of an information product
with IP-MAP. In Proceedings of the Int’l Conference
on Information Quality, pages 1–16.
DataQualityEvaluationofScientificDatasets-ACaseStudyinaPolicySupportContext
173
Shanks, G. and Corbitt, B. (1999). Understanding data qual-
ity: Social and cultural aspects. In Proceedings of
the Australasian Conference on Information Systems,
pages 785–797.
Song, S. and Chen, L. (2011). Differential dependen-
cies: Reasoning and discovery. ACM Transactions on
Database Systems, 26(3), 16.
Szlichta, J., Godfrey, P., and Gryz, J. (2012). Fundamen-
tals of order dependencies. Proceedings of the VLDB
Endowment, 5(11):1120–1231.
van der Vlist, E. (2007). Schematron. O’Reilly Media.
Vassiliadis, P. (2009). A survey of Extract-Transform-Load
technology. Int’l Journal of Data Warehousing &
Mining, 5(3):1–27.
APPENDIX
IP-MAP
The IP-MAP graphical model is aimed at creating
a systematic representation for capturing the details
associated with the manufacturing of an informa-
tion product. An information product is produced
by means of processing activities and data quality
checks on raw data and semi-processed information.
Eight construct blocks are the main components of an
IP-MAP: source block, customer block, data quality
block, processing block, decision block, data storage
block, organizational boundary block, and informa-
tion system boundary block. Each construct block is
identified by a unique name and is further described
by a set of attributes (i.e., metadata). Figure 2 lists the
symbols assigned to each construct block.
Document Type Definition
<!DOCTYPE rules [
<!ELEMENT rules (rule_definition+)>
<!ELEMENT rule_definition (table_name+,
(rule_cr|rule_fd|rule_od|
rule_dd|rule_ec|rule_cc))>
<!ELEMENT rule_cr (if, then)>
<!ELEMENT rule_fd (lhs, rhs, when?)>
<!ELEMENT rule_od (lhs, rhs, when?,
partition_on?)>
<!ELEMENT rule_dd (lhs_dd, rhs_dd, when?,
partition_on?)>
<!ELEMENT rule_ec ((column+|(lhs_ec, rhs_ec)),
when?, join_on?)>
<!ELEMENT rule_cc (check, join_on?)>
<!ELEMENT if (condition|conditions)>
<!ELEMENT then (condition|conditions)>
<!ELEMENT when (condition|conditions)>
<!ELEMENT check (condition|conditions)>
<!ELEMENT join_on (column+)>
<!ELEMENT lhs (column_name+)>
<!ELEMENT rhs (column_name+)>
Construct name Construct symbol
Source (input)
Block
Customer (output)
Block
Data Quality
(evaluation) Block
Processing
Block
Decision
Block
Data Storage
Block
Organizational
Boundary Block
Information System
Boundary Block
Figure 2: IP-MAP construct symbols.
<!ELEMENT lhs_ec (column)>
<!ELEMENT rhs_ec (column|(table_name?,
disj_attr+))>
<!ELEMENT lhs_dd (distance_condition+)>
<!ELEMENT rhs_dd (distance_condition+)>
<!ELEMENT partition_on (column_name+)>
<!ELEMENT column (table_name?, column_name)>
<!ELEMENT disj_attr (column_name+)>
<!ELEMENT distance_condition (column_name,
comparison_operator, constant)>
<!ELEMENT conditions ((condition|conditions)+,
boolean_operator)>
<!ELEMENT condition (lside,
comparison_operator, rside)>
<!ELEMENT lside (column|
arithmetic_computation)>
<!ELEMENT rside (constant|column|
arithmetic_computation)>
<!ELEMENT arithmetic_computation ((constant|
column|arithmetic_computation)+,
arithmetic_operator)>
<!ELEMENT table_name (#PCDATA)>
<!ELEMENT column_name (#PCDATA)>
<!ELEMENT comparison_operator (#PCDATA)>
<!ELEMENT boolean_operator (#PCDATA)>
<!ELEMENT arithmetic_operator (#PCDATA)>
<!ELEMENT constant (#PCDATA)>
<!ATTLIST rule_definition name CDATA #IMPLIED
type (conditional_rule|functional_dependency|
order_dependency|distance_dependency|
existence_constraint|check_constraint)
#REQUIRED>
<!ATTLIST rule_od type (direct|inverse)
#REQUIRED>
<!ATTLIST rule_ec type (ec_dep|ec_bidir|
ec_disj|ec_attr) #REQUIRED> ]>
DATA2013-2ndInternationalConferenceonDataManagementTechnologiesandApplications
174
