Analyzing Quality Criteria in Role-based Identity and Access
Management
Michael Kunz
1
, Ludwig Fuchs
1,2
, Michael Netter
1,2
and G
¨
unther Pernul
1
1
University of Regensburg, Regensburg, Germany
2
Nexis GmbH, M
¨
unster, Germany
Keywords:
Role Quality, Role Mining, RBAC, Identity and Access Management.
Abstract:
Roles have turned into the de facto standard for access control in enterprise identity management systems.
However, as roles evolve over time, companies struggle to develop and maintain a consistent role model.
Up to now, the core challenge of measuring the current quality of a role model and selecting criteria for its
optimization remains unsolved. In this paper, we conduct a survey of existing role mining techniques and
identify quality criteria inherently used by these approaches. This guides organizations during the selection of
a role mining technique that matches their company-specific quality preferences. Moreover, our analysis aims
to stimulate the research community to integrate quality metrics in future role mining approaches.
1 INTRODUCTION
Regulating access to resources is an elementary func-
tion of every identity management system (IdMS).
Not just as a result of governmental regulations or
compliance requirements like the Sarbanes-Oxley Act
((SOX, 2002)), Basel III ((Basel Comittee on Banking
Supervisions, 2010)), or the EU General Data Protec-
tion Regulation ((European Union, 2012)) in its re-
vised form, especially medium- and large-sized com-
panies are forced to control access to sensitive infor-
mation. Over the past decades, Role-Based Access
Control (RBAC, (Sandhu et al., 1996)), has become
the de facto standard for managing access to resources
in IT systems. In RBAC, roles act as intermediary be-
tween users and permissions. Despite being widely
used, RBAC struggles with the dynamic evolution of
role models over time. Besides the daily user ad-
ministration, the central challenge after setting up a
role model is its strategic maintenance. Role system
maintenance focuses on updating and cleansing role
configurations, discarding unused, and defining new
roles. Changing business processes, organizational
structures, or security policies and newly imposed
regulations force administrators to quickly adapt the
access control structures in place. Commonly, this
leads to an increasing role number, an overall reduc-
tion of the role model quality and the advent of se-
curity vulnerabilities due to wrongly assigned or out-
dated role definitions.
In order to mitigate the risk of increasing security
vulnerabilities in RBAC, one cornerstone of ensuring
a high role model quality is the periodic assessment
of the role model components, such as the user-role
assignments (UA), the role-permission assignments
(RA), or role hierarchy structures. Role mining ap-
proaches that support organizations during their ini-
tial setup of an RBAC model have attracted the at-
tention of researchers. Over the last four years, for
instance, a variety of research groups have published
approaches to generate an initial set of roles. However
hardly any attention has been drawn to the mainte-
nance of existing role models. Recently, the need for
investigating and cleaning role model structures has
been highlighted by (Fuchs et al., 2014). However,
the core challenge of measuring the current quality of
a role model and selecting criteria for its optimization
still remains unsolved. Due to its importance for ac-
cess control it is likely that role mining is re-applied
periodically by organizations in order to ensure role
model correctness. Our work builds on both, the in-
depth investigation of the research area as well as
our practical experience from industry projects within
medium- and large-sized companies dealing with the
setup and management of a role-based IdMS. Simi-
lar to (Fuchs et al., 2014), we argue that the practical
project requirements cannot be considered to a suffi-
cient extent by the available role mining approaches.
In particular, we address the following research ques-
tion (RQ): Which quality criteria are employed in ex-
64
Kunz M., Fuchs L., Netter M. and Pernul G..
Analyzing Quality Criteria in Role-based Identity and Access Management.
DOI: 10.5220/0005232100640072
In Proceedings of the 1st International Conference on Information Systems Security and Privacy (ICISSP-2015), pages 64-72
ISBN: 978-989-758-081-9
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
Use of quality criteria in existing
Role Mining approaches
1
Literature
Survey
2
Classification &
Criteria elicitation
RQ
Role
Mining
catalogue
Figure 1: Research methodology.
isting role mining approaches?
Hence, the contribution of this work is twofold.
In order to close the existing research gap, this paper
firstly analyzes the development of role mining, pre-
senting a survey of the field and underlining the rapid
development in the area. In this respect, it builds on
an existing survey of the area published in the year
2011. During execution, various role mining algo-
rithms rely on some sort of quality criteria to differ-
ent extents. As a result, this paper secondly extracts
potential criteria for rating the quality of role models
included in current role mining approaches. It points
out the differences among the approaches and under-
lines the need for a structured quality rating process.
We thereby aim at stimulating the research commu-
nity and encourage them to enrich existing role min-
ing approaches considering quality criteria in a struc-
tured manner.
2 RELATED WORK
In todays medium- and large-sized companies, Role-
based Access Control has become the de facto stan-
dard for controlling user access to resources. As a
result of the large amount of research output, sev-
eral surveys of the general area of roles in IT secu-
rity have been presented (e.g. (Zhu and Zhou, 2008)
and (Fuchs et al., 2011)). Authors lately agreed upon
the growing importance of role development in gen-
eral and automated role mining in specific. Fuchs et
al., for instance, provided an evaluation of role de-
velopment approaches in (Fuchs and M
¨
uller, 2009)
and (Fuchs and Meier, 2011). Since their publication,
the research output in the area has grown more than
double, requiring a survey update in order to give an
in-depth understanding of recent developments.
During the initial setup of a role model, role min-
ing algorithms inherently rely on different quality cri-
teria to various extents. Nevertheless, no structured
analysis considering those criteria has been executed
so far. It has rather been shown that popular role
mining approaches like (Molloy et al., 2010), (Giblin
et al., 2010), or (Takabi and Joshi, 2010) do not offer
the guidance required to judge the correctness of role
definitions and role models (Fuchs et al., 2014). In
practice, however, it is very likely that companies re-
apply role mining techniques in order to ensure role
model correctness after having employed a role min-
ing approach for the initial role setup. Yet, none of
the related work in the field focuses on quality criteria
applied during role system maintenance. In (Molloy
et al., 2008), the authors investigated the usage of se-
lected metrics like the Weighted Structural Complex-
ity (WSC) for analyzing role system states. (Fuchs
and M
¨
uller, 2009) described mechanisms for periodi-
cally evaluating a role systems quality but do not con-
sider the scalability of their approach in large real-
world scenarios. (Fuchs et al., 2014) recently pro-
posed the integration of a distinct quality rating and
role classification phase in their role optimization pro-
cess model. However, they do not present an overview
of available quality criteria and their application. As
a result, the core challenge of measuring the current
quality of a role model still remains unsolved.
3 METHODOLOGY
Our methodology to answer the research questions
presented in Section 1 is depicted in Figure 1. At
first, we survey the research area and create a catalog
of role mining approaches that serves as the basis for
further analyses (step 1). For this literature survey, we
follow the methodology proposed by (Levy and Ellis,
2006). We carried out a bibliographic database search
including the ACM Digital Library, DBLP, IEEE Dig-
ital Library, and Google Scholar using the keyword
”role mining”. To arrive at a complete catalog of role
mining approaches, author and reference search for
each publication was applied to identify previously
undiscovered research. Finally, papers that do not
present role mining techniques were removed from
the catalog (e.g. (Molloy et al., 2008)). Consecu-
tively, we classify recent role mining publications ac-
cording to the scheme presented by (Fuchs and Meier,
2011). Additional clusters were added for role min-
ing approaches that used new techniques. Note, that
approaches that rely on more than one role mining
technique were clustered based on the predominant
technique being used. At this stage we completed our
first contribution by extending the existing survey and
highlighting the rapid increase of research within the
last three years as well as the further diversification
of role mining techniques used. During step 2 of our
methodology we answer RQ by investigating the us-
age of quality criteria in role mining algorithms.
4 ROLE MINING SURVEY
This section extends a survey on role mining research
conducted by (Fuchs and Meier, 2011). Figure 2 un-
AnalyzingQualityCriteriainRole-basedIdentityandAccessManagement
65
1
0
1 1
3
8
9
13 13
21 21
0
4
8
12
16
20
24
Number of Publications
Year of Publication
Figure 2: Development of the research area.
derlines the importance of an updated interpretation
of role detection approaches due to the increase in re-
searchers’ attention during the last three years. Be-
tween 2011 and 2013, 55 papers related to role mining
have been published, representing an increase of 141
percent compared to the number of publications from
2003 - 2010 (39). While role mining in general con-
sists of a pre-processing phase, a role detection phase,
and a post-processing phase (Fuchs and Meier, 2011),
the remainder of this survey focuses on the role detec-
tion phase as the core element of every role mining al-
gorithm. During this phase, suitable roles are created
based on an existing set of user-permission assign-
ments (UPA). The algorithms can be grouped accord-
ing to the underlying technique. While the first six
techniques have been previously introduced in (Fuchs
and Meier, 2011), we identified three additional tech-
niques (Visual Role Mining, Boolean Matrix Decom-
position, and Attribute-based Role Mining) being ap-
plied to solve the role mining problem for the first
time. Additionally to the 21 algorithms published in
(Fuchs and Meier, 2011), 26 out of the 55 approaches
from 2011 to 2013 have been categorized, while the
rest is related to either pre- or post-processing phase.
Subsequently, each technique and representative pub-
lications are presented:
Subset Enumeration aims to discover roles
through creating all possible intersections of permis-
sion sets. Due to the exponential complexity of enu-
merating all possible subsets, algorithms such as (Xu
and Stoller, 2013b) use heuristics for role candidate
selection. Hence, role mining algorithms based on
this technique strive to balance complexity and qual-
ity of results. With 13 available algorithms, this is the
most frequently used technique in role mining.
Clustering is a role mining approach that is di-
rectly derived from data mining (Frank et al., 2012).
Using a UPA-matrix as input, this technique searches
for clusters with similar permissions. However, clus-
tering approaches struggle with limitations such as re-
quiring users or permissions to be only part of one
single cluster (Frank et al., 2012). To solve these
challenges, several clustering variants exist. Exam-
ples include iterative application of the technique or
reduction of the input (Lu et al., 2013; Huang et al.,
2012).
Graph Optimization uses bipartite graphs to rep-
resent the UPA (Colantonio et al., 2009). As roles rep-
resent an intermediary between the two disjoint ver-
tices of users and permissions, those approaches aim
at converting the bipartite into a tripartite graph or a
representation with sub-graphs. Roles are then repre-
sented by the middle vertex (Colantonio et al., 2009)
or each necessary sub-graph (Mandala et al., 2012).
Frequent Permission Set Mining has its roots in
marketing analysis and algorithms of generic frequent
set mining (Agrawal et al., 1993). Initially used for
the study of consumers’ purchase behavior, it is ap-
plied by role mining algorithms in order to discover
permissions that frequently occur together. The pre-
sumption is that permission sets that appear together
can be interpreted as role candidates.
Formal Concept Analysis is a data mining tech-
nique similar to clustering, overcoming the limitation
of only assigning entities to one group (Molloy et al.,
2008). Related algorithms build a concept lattice from
the UPA-matrix and reduce duplicate information. A
concept lattice is a construct similar to a graph and
represents roles and their connection in a partially
ordered collection of clusters which again consist of
permissions and users.
Heuristic Matrix Selection is similar to Subset
Enumeration without the initial role set being based
on intersections of permissions. Instead, it iterates
through rows (or columns) and picks role candidates
successively according to the highest number of given
assignments (e.g permissions assigned to a user). Ini-
tially, each row/column is treated as a role. Subse-
quently, a cross-check for duplicate roles is conducted
(Blundo and Cimato, 2010).
Visual Role Mining is a fairly new technique ini-
tially proposed in (Colantonio et al., 2012). It re-
orders rows and columns of the input UPA-matrix in
order to create clusters of adjoined permissions. Dis-
played to an administrator, the underlying assump-
tion is that humans’ cognitive capabilities and context
knowledge are better suited to discover proper roles
compared to purely algorithm-based approaches.
Boolean Matrix Decomposition is an approach
that directly addresses the Role Mining Problem
(RMP) introduced by (Vaidya et al., 2007). This for-
mal definition of role mining and targets at decom-
posing the boolean UPA-matrix into two separate ma-
trices, a UA-matrix and a PA-matrix. By dividing the
initial UPA-matrix into two consistent sub-matrices,
the columns of the UA-matrix and the rows of the PA-
matrix build up the set of roles.
ICISSP2015-1stInternationalConferenceonInformationSystemsSecurityandPrivacy
66
Attribute-based Role Mining such as (Frank
et al., 2009) are trying to incorporate business in-
formation through attributes into role mining. They
rely on the assumption that additional semantic data
is available and can be taken into account. Attribute-
based approaches combine other techniques and en-
rich them with attribute-based mechanisms to arrive
at an improved role set.
5 QUALITY-RELATED CRITERIA
After their classification we examined role mining al-
gorithms regarding their decision making processes
of including certain role candidates in their final out-
put. We argue that this central decision making pro-
vides well-suited indicators for quality management
in RBAC. This assumption is based on the claims
of several publications (e.g. (Xu and Stoller, 2012),
(Zhang et al., 2013b)) of outperforming competitive
approaches in terms of the quality of generated roles.
A total of 23 different quality criteria can be identi-
fied. They either focus on the quality of the overall
RBAC state, the quality of single roles, or both. At
first we focus on RBAC state quality criteria. Sec-
ondly, we examine criteria that deal with the quality
of an individual role (cf. Table 1). Note that for some
criteria (e.g. Exclude Unused Permissions) additional
input information is required. In the following, we
present a detailed interpretation of the quality criteria
and group them according to their focus.
Achieve Completeness. Completeness refers to the
exact representation of the original access control
state, i.e. the goal is to cover the initial U PA-matrix
with the resulting set of roles. In contrast to most
approaches (e.g. (Zhang et al., 2007; Vaidya et al.,
2006; Blundo and Cimato, 2010)), some techniques
allow to deviate to a certain extent from the initial
UPA-matrix based on a given threshold (the so called
δ RMP (Vaidya et al., 2007))(Chu et al., 2012; Lu
et al., 2008). Completeness therefore measures the
quality of a RBAC state by measuring the degree to
which a resulting role set represents the initial access
control situation.
Reduce Number of Roles. Initially formulated in
the RMP, the goal of having as few roles as possible is
based on the assumption that complexity of RBAC is
directly connected to the number of roles maintained.
Thus, the number of roles in a given RBAC state is a
quality criteria usable to rate the estimated adminis-
trative efforts to manage the role model. Depending
on the size of a company in terms of its employees,
permissions and UPA, this measure can be normal-
ized in order to allow for a comparison of role models
in different organizations.
Decrease Role Set Similarity. Quality criteria re-
lated to Role Set Similarity measure the distance be-
tween two given sets of roles. They are mainly used
for the measurement of the dissimilarity of RBAC
states (e.g. in (Zhang et al., 2013a)) or the difference
between a current RBAC state and a targeted state. In
(Jafari et al., 2009), for instance, the permission simi-
larity is measured using the Euclidean Distance. Fur-
thermore, the Jaccard Similarity is a popular metric
used in a variety of approaches (Mandala et al., 2012;
Xu and Stoller, 2013a; Chu et al., 2012).
Minimize Users/Permissions per Role & Min-
imize/Maximize Roles per User/Permission.
Quality criteria assigned to this category target at
two main objectives. First, the definition of an upper
bound per role limiting the amount of users assigned
to one role. Second, the objective of finding as few
as possible permissions that are placed in a role. This
can, for instance, be applied in case an organization
aims at defining a larger number of small roles
for employees that exactly fit their specialist tasks.
On the contrary, maximizing the number of users or
permissions per role can be beneficial other scenarios,
e.g. when organizations aim at defining a small set of
large roles. Moreover, Minimizing/Maximizing the
roles per element (user or permission) is applied in
seven existing role mining approaches (e.g. see (Lu
et al., 2013; Ma et al., 2013)). This can be useful in
order to minimize the role model complexity in terms
of relationships among the role model elements.
Fullfill Role Constraints. Role Constraints impose
restrictions on the definition of roles. For instance,
(Ma et al., 2012) consider Segregation of Duty (SOD)
policies that entail mutually exclusive permissions in
the RBAC state. Other policies, such as the four-eye-
principle, that affect the user assignments of a role are
also conceivable.
Reduce WSC. In contrast to most other quality cri-
teria the so called Weighted Structural Complexity
(WSC) is a widely-used heuristic to rate the complex-
ity of a RBAC model. Originally introduced by (Mol-
loy et al., 2008), it applies weights to different opti-
mization objectives. It can be seen as one of the most
advanced measures that solely relies on the compo-
nents of an RBAC state. It is usable for both, indi-
vidual roles and role sets, and thus allows for a good
AnalyzingQualityCriteriainRole-basedIdentityandAccessManagement
67
Table 1: Quality Criteria in existing approaches.
Quality criteria State Individual Role State + Individual Role
Technique / Focus Paper
Achieve Completeness
Reduce Number of Roles g
Decrease Role Set Similarity
Minimize Users per Role
Maximize Users per Role
Minimize Roles per User
Minimize Roles per Permission
Minimize Permissions per Role
Maximize Permissions per Role
Fulfill Role Constraints
Reduce WSC
Optimize Matrix Sorting
Decrease Permission Similarity
Reduce Role Redundancy
Increase User Similarity
Decrease Permission Attr. Sim.
Increase Role Coverage
Exclude Unused Permissions
Consider Timestamp
Consider Role Attributes
Consider User Attributes
Group by Attributes
Subset Enumeration
(Chu et al., 2012)
(Colantonio et al., 2008a)
(Huang et al., 2012)
(Lu et al., 2013)
(Mitra et al., 2013)
(Molloy et al., 2012)
(Vaidya et al., 2006)
(Vaidya et al., 2007)
(Vaidya et al., 2010b)
(Vaidya et al., 2010a)
(Xu and Stoller, 2012)
(Xu and Stoller, 2013b)
(Zhang et al., 2013a)
Clustering
(Frank et al., 2008)
(Frank et al., 2012)
(Frank et al., 2013)
(Kumar et al., 2011)
(Schlegelmilch and Steffens, 2005)
Graph Optimization
(Colantonio et al., 2009)
(Colantonio et al., 2010)
(Ene et al., 2008)
(Gal-Oz et al., 2011)
(Hingankar and Sural, 2011)
(Mandala et al., 2012)
(Zhang et al., 2007)
Frequent Permission Set Mining
(Colantonio et al., 2008b)
(Jafari et al., 2009)
(Ma et al., 2010)
(Ma et al., 2012)
(Ma et al., 2013)
(Zhang et al., 2008)
Formal Concept Analysis
(Molloy et al., 2008)
(Wang et al., 2012)
(Wong et al., 2012)
Heuristic Matrix Selection
(Blundo and Cimato, 2010)
(Blundo and Cimato, 2013)
(Huang et al., 2010)
Visual Role Mining
(Colantonio et al., 2012)
(Eucharista, 2013)
Boolean Matrix Decomposition
(John et al., 2012)
(Lu et al., 2012)
(Lu et al., 2008)
(Uzun et al., 2011)
(Ye et al., 2013)
Attribute-based Approaches
(Frank et al., 2009)
(Han et al., 2012)
(Li et al., 2012)
Quality Criteria State Individual Role State + Individual Role
Technique / Focus Paper
Achieve Completeness
Reduce Number of Roles g
Decrease Role Set Similarity
Minimize Users per Role
Maximize Users per Role
Minimize Roles per User
Minimize Roles per Permission
Minimize Permissions per Role
Maximize Permissions per Role
Fulfill Role Constraints
Reduce WSC
Optimize Matrix Sorting
Decrease Permission Similarity
Reduce Role Redundancy
Increase User Similarity
Decrease Permission Attr. Sim.
Increase Role Coverage
Exclude Unused Permissions
Consider Timestamp
Consider Role Attributes
Consider User Attributes
Group by Attributes
Subset Enumeration
(Chu et al., 2012)
(Colantonio et al., 2008a)
(Huang et al., 2012)
(Lu et al., 2013)
(Mitra et al., 2013)
(Molloy et al., 2012)
(Vaidya et al., 2006)
(Vaidya et al., 2007)
(Vaidya et al., 2010b)
(Vaidya et al., 2010a)
(Xu and Stoller, 2012)
(Xu and Stoller, 2013b)
(Zhang et al., 2013a)
Clustering
(Frank et al., 2008)
(Frank et al., 2012)
(Frank et al., 2013)
(Kumar et al., 2011)
(Schlegelmilch and Steffens, 2005)
Graph Optimization
(Colantonio et al., 2009)
(Colantonio et al., 2010)
(Ene et al., 2008)
(Gal-Oz et al., 2011)
(Hingankar and Sural, 2011)
(Mandala et al., 2012)
(Zhang et al., 2007)
Frequent Permission Set Mining
(Colantonio et al., 2008b)
(Jafari et al., 2009)
(Ma et al., 2010)
(Ma et al., 2012)
(Ma et al., 2013)
(Zhang et al., 2008)
Formal Concept Analysis
(Molloy et al., 2008)
(Wang et al., 2012)
(Wong et al., 2012)
Heuristic Matrix Selection
(Blundo and Cimato, 2010)
(Blundo and Cimato, 2013)
(Huang et al., 2010)
Visual Role Mining
(Colantonio et al., 2012)
(Eucharista, 2013)
Boolean Matrix Decomposition
(John et al., 2012)
(Lu et al., 2012)
(Lu et al., 2008)
(Uzun et al., 2011)
(Ye et al., 2013)
Attribute-based Approaches
(Frank et al., 2009)
(Han et al., 2012)
(Li et al., 2012)
Uses criteria: Yes No
comparability of RBAC states. As a result of its pop-
ularity, several existing role mining approaches are
able to consider the WSC.
Optimize Matrix Sorting. Matrix sorting aims at
covering an initial access control state by sorting the
input UPA-matrix based on user accounts with sim-
ilar permissions and permissions that are assigned
a similar set of user accounts. (Colantonio et al.,
2012) introduced the ADVISER and EXTRACT al-
gorithms that generate a matrix representation of the
initial UPA-matrix that clusters permissions and user
accounts together. As a result, large areas covering
initial UPA can be visually detected by a human role
engineer.
Similarities & Redundancy. Well-known similar-
ity metrics can be applied to the various elements of
a RBAC state in order to measure its quality. (Takabi
and Joshi, 2010) gives an overview of possible appli-
cations of the Jaccard Similarity in the context of role
management. He discusses three similarity metrics
that can be applied on the assignment types of a role
(user, permission and role hierarchy). They can fur-
ther be used to compute the similarity of two role sets
(cf. Decrease Role Set Similarity). Besides examin-
ing the similarity of assignment types of a role, simi-
larity metrics are applied to attributes of role compo-
ICISSP2015-1stInternationalConferenceonInformationSystemsSecurityandPrivacy
68
nents. They can, for instance, be used to create a role
set based on the location attribute of all user accounts.
Distance measures are applied to identify redundant
roles (Chu et al., 2012).
Increase Role Coverage. The Role Coverage is
formally defined in (Zhang et al., 2008) as the frac-
tion of role-covered UPA by the initial U PA. Compa-
nies aim at achieving a high role coverage in order to
foster the benefits of RBAC compared to other access
control models. The implication of reducing admin-
istrative costs through RBAC is represented through
this criterion.
Attribute-related Criteria. Attribute-related crite-
ria evaluate the quality of a role based on its attributes
or attributes of its components. Permission usage de-
rived from access logs, for example, can be used to
display the actual usage of privileges by employees.
It offer insights into unused PA that can potentially be
removed during the next refinement of a role (Molloy
et al., 2012). Furthermore, restrictions on the compo-
sition of a role, e.g. by allowing only certain attributes
of users in a role are possible (Xu and Stoller, 2012).
6 DISCUSSION
This work is motivated by the gap between the re-
cent uprising of role mining and the practical need for
periodic quality assessment of the resulting role mod-
els. The presented survey underlined the significant
growth (141%) of published papers in the recent past.
We have shown that every role mining approach re-
lies on one or more quality criteria, mostly implicitly
without providing a structured integration of quality
management. In the following we present a short dis-
cussion of our quality-related findings from Table 1.
Firstly, it can be seen that the main quality crite-
rion in role mining is to arrive at an exact representa-
tion of existing access control states. This criterion is
– to a varying extent – considered by all available ap-
proaches, except for (Jafari et al., 2009) which derive
the roles solely from access history logs. Secondly,
Table 1 shows that a large number of approaches fo-
cus on generating as few roles as possible (Reduce
Number of Roles). Interestingly, as the WSC is a
potential criterion which is able to represent this and
other measures (by modifying its weight factors), re-
cent approaches try to use this metric as a heuristic for
producing high quality roles (Xu and Stoller, 2013b;
Eucharista, 2013; Ye et al., 2013). This can be inter-
preted as an indicator that research is already trying
to integrate sophisticated measures into role mining.
Other interesting results are, that criteria with
practical relevance up to now are only considered by
few existing approaches. Timestamps as an attribute
of permissions are, for instance, only considered by
one approach (Mitra et al., 2013). However, their in-
tegration into role mining seems promising as they
heavily can influence role design. Sets of permis-
sions activated together within a certain period of time
can e.g. represent good candidate permissions for a
role. We furthermore noted that several quality crite-
ria well-known in practice have not yet been included
in any role mining approach at all. This includes cri-
teria like the maximum allowed number of roles in
a role model, role usage or hierarchy restrictions. It
seems straightforward to integrate a maximum thresh-
old of roles to be found through automatic role dis-
covery in order to ensure the maintainability of the
whole role set. Intuitively, result sets can always be
limited by just taking the desired number of roles af-
ter sorting them according to a predefined criterion.
Furthermore, the usage of UA over a certain period
of time (i.e. the activation of roles) can hint at out-
dated role definitions. Several approaches are able to
take existing roles into consideration (e.g. (Molloy
et al., 2008)) but do not integrate usage data. More-
over, restrictions on the hierarchy of a RBAC state can
represent one way to reduce complexity and increase
the quality of either a single role or the whole RBAC
state. In practice, deployed RBAC states feature un-
limited depth, sometimes even resulting in hierarchy
loops. Limiting the maximum allowed number of par-
ent or child roles of a role or the maximal hierarchy
depth can ease administrative staffs understanding of
the overall role model. Several post-processing ap-
proaches already outline the need for an inspection of
the role hierarchy (e.g. (Guo et al., 2008; Takabi and
Joshi, 2010)). Yet, they focus on removing duplicate
hierarchy depiction and finding minimal hierarchical
assignments, not on cardinality restrictions.
7 CONCLUSION
Role-Based Access Control as the de facto standard
for managing access privileges in organizations strug-
gles with the dynamic evolution of role models over
time. As a result the quality of RBAC states initially
modeled using role mining techniques decreases over
time. In order to address this challenge, role mining
mechanisms applied during role system maintenance
need to be extended in order to integrate a dedicated
quality management stage for rating and improving
a role system state on the basis of company-specific
quality criteria. In this paper we presented two con-
AnalyzingQualityCriteriainRole-basedIdentityandAccessManagement
69
tributions in that respect. We firstly provided a survey
giving an overview of current role mining approaches.
The significant increase of research activity and the
growing number of applied techniques for generating
roles during the last three years underlines the rele-
vance and diversification of the area. By extracting
criteria that are dedicated to improving role quality
from currently available role mining approaches we
were able to answer the research question stated in
Section 1. We have shown on which quality crite-
ria current role mining approaches rely and revealed
a number of practically relevant but yet untreated cri-
teria in research. The integration of quality mecha-
nisms in order to allow for an improved selection of
role mining approaches in a given scenario based on
company-specific quality criteria.
ACKNOWLEDGEMENTS
The research leading to these results was supported
by the “Bavarian State Ministry of Education, Sci-
ence and the Arts” as part of the FORSEC research
association. This work would not have been possible
without our student Christian Wawarta.
REFERENCES
Agrawal, R., Imieli
´
nski, T., and Swami, A. (1993). Min-
ing association rules between sets of items in large
databases. In SIGMOD Record, volume 22, pages
207–216. ACM.
Basel Comittee on Banking Supervisions (2010). Basel III:
Int. framework for liquidity risk measurement, stan-
dards and monitoring.
Blundo, C. and Cimato, S. (2010). A simple role mining al-
gorithm. In Proc. of the 2010 Symp. on Applied Com-
puting. ACM.
Blundo, C. and Cimato, S. (2013). Constrained role mining.
In 6th Int. Workshop on Security and Trust Manage-
ment, pages 289–304. Springer.
Chu, V. W., Wong, R. K., and Chi, C.-H. (2012). Over-
fitting and error detection for online role mining. Int.
Journal of Web Services Research, 9(4):1–23.
Colantonio, A., Di Pietro, R., and Ocello, A. (2008a). A
cost-driven approach to role engineering. In Proc. of
the 2008 Symp. on Applied Computing. ACM.
Colantonio, A., Di Pietro, R., and Ocello, A. (2008b).
Leveraging lattices to improve role mining. In Proc. of
The IFIP TC 11 23rd Int. Information Security Conf.
Springer.
Colantonio, A., Di Pietro, R., Ocello, A., and Verde, N. V.
(2009). A probabilistic bound on the basic role min-
ing problem and its applications. In Emerging Chal-
lenges for Security, Privacy and Trust, pages 376–386.
Springer.
Colantonio, A., Di Pietro, R., Ocello, A., and Verde, N. V.
(2010). Taming role mining complexity in rbac. Com-
puters & Security, 29(5):548–564.
Colantonio, A., Di Pietro, R., Ocello, A., and Verde, N. V.
(2012). Visual role mining: A picture is worth a thou-
sand roles. IEEE Transactions on Knowledge and
Data Engineering, 24(6):1120–1133.
Ene, A., Horne, W., Milosavljevic, N., Rao, P., Schreiber,
R., and Tarjan, R. E. (2008). Fast exact and heuristic
methods for role minimization problems. In Proc. of
the 13th Symp. on Access Control Models and Tech-
nologies. ACM.
Eucharista, A.and Haribaskar, K. (2013). Visual elicitation
of roles: using a hybrid approach. Oriental Journal of
Computer Science & Technology, 6(1):103–110.
European Union (2012). General data protection regulation.
Frank, M., Basin, D., and Buhmann, J. M. (2008). A class
of probabilistic models for role engineering. In Proc.
of the 15th ACM Conf. on Computer and Communica-
tions Security. ACM.
Frank, M., Buhman, J. M., and Basin, D. (2013). Role min-
ing with probabilistic models. ACM Transactions on
Information and System Security, 15(4):15:1–15:28.
Frank, M., Streich, A. P., Basin, D., and Buhmann, J. M.
(2009). A probabilistic approach to hybrid role min-
ing. In Proc. of the 16th ACM Conf. on Computer and
communications security, pages 101–111. ACM.
Frank, M., Streich, A. P., Basin, D., and Buhmann, J. M.
(2012). Multi-assignment clustering for boolean data.
Journal of Machine Learning Research, 13(1):459–
489.
Fuchs, L., Kunz, M., and Pernul, G. (2014). Role model op-
timization for secure role-based identity management.
In Proc. of the 22nd European Conf. on Information
Systems.
Fuchs, L. and Meier, S. (2011). The role mining process
model - underlining the need for a comprehensive re-
search perspective. In Proc. of the 6th Int. Conf. on
Availability, Reliability and Security. IEEE.
Fuchs, L. and M
¨
uller, C. (2009). Automating periodic
role-checks: A tool-based approach. In Business
Services: Konzepte, Technologien, Anwendungen:
9. Int.e Tagung Wirtschaftsinformatik, volume 246.
OCG, Wien.
Fuchs, L., Pernul, G., and Sandhu, R. (2011). Roles in in-
formation security–a survey and classification of the
research area. Computers & Security, 30(8):748–769.
Gal-Oz, N., Gonen, Y., Yahalom, R., Gudes, E., Rozenberg,
B., and Shmueli, E. (2011). Mining roles from web
application usage patterns. In Trust, Privacy and Secu-
rity in Digital Business, volume 6863 of Lecture Notes
in Computer Science, pages 125–137. Springe.
Giblin, C., Graf, M., Karjoth, G., Wespi, A., Molloy, I.,
Lobo, J., and Calo, S. B. (2010). Towards an inte-
grated approach to role engineering. In SafeConfig,
pages 63–70. ACM.
Guo, Q., Vaidya, J., and Atluri, V. (2008). The role hi-
erarchy mining problem: Discovery of optimal role
hierarchies. In Computer Security Applications Conf.
IEEE.
ICISSP2015-1stInternationalConferenceonInformationSystemsSecurityandPrivacy
70
Han, D.-j., Zhuo, H.-k., Xia, L.-t., and Li, L. (2012). Per-
mission and role automatic assigning of user in role-
based access control. Journal of Central South Uni-
versity, 19:1049–1056.
Hingankar, M. and Sural, S. (2011). Towards role min-
ing with restricted user-role assignment. In 2nd Int.
Conf. on Wireless Communication, Vehicular Tech-
nology, Information Theory and Aerospace Electronic
Systems Technology.
Huang, C., Sun, J.-l., Wang, X.-y., and Si, Y.-j. (2010).
Minimal role mining method for web service compo-
sition. Journal of Zhejiang University SCIENCE C,
11(5):328–339.
Huang, H., Shang, F., and Zhang, J. (2012). Approximation
algorithms for minimizing the number of roles and ad-
ministrative assignments in rbac. In 36th Annual Com-
puter Software and Applications Conf. Workshops.
IEEE.
Jafari, M., Chinaei, A., Barker, K., and Fathian, M. (2009).
Role mining in access history logs. Journal of Infor-
mation Assurance and Security, 38.
John, J., Sural, S., Atluri, V., and Vaidya, J. (2012). Role
mining under role-usage cardinality constraint. In In-
formation Security and Privacy Research, volume 376
of IFIP Advances in Information and Communication
Technology, pages 150–161. Springer.
Kumar, R., Sural, S., and Gupta, A. (2011). Mining rbac
roles under cardinality constraint. In Information Sys-
tems Security, pages 171–185. Springer.
Levy, Y. and Ellis, T. J. (2006). A Systems Approach to
Conduct an Effective Literature Review in Support
of Information Systems Research. Informing Science
Journal, 9:181–212.
Li, R., Wang, W., Ma, X., Gu, X., and Wen, K. (2012). Min-
ing roles using attributes of permissions. Int. Jour-
nal of Innovative Computing, Information and Con-
trol, 8(11):7909–7924.
Lu, H., Hong, Y., Yang, Y., Duan, L., and Badar, N. (2013).
Towards user-oriented rbac model. In Data and Ap-
plications Security and Privacy XXVII, volume 7964
of Lecture Notes in Computer Science, pages 81–96.
Springer.
Lu, H., Vaidya, J., and Atluri, V. (2008). Optimal boolean
matrix decomposition: Application to role engineer-
ing. In Proc. of the 24th IEEE Int. Conf. on Data En-
gineering. IEEE.
Lu, H., Vaidya, J., Atluri, V., and Hong, Y. (2012).
Constraint-aware role mining via extended boolean
matrix decomposition. IEEE Transactions on De-
pendable and Secure Computing, 9(5):655–669.
Ma, X., Li, R., and Lu, Z. (2010). Role mining based on
weights. In Proc. of the 15th Symp. on Access Control
Models and Technologies. ACM.
Ma, X., Li, R., Lu, Z., and Wang, W. (2012). Mining con-
straints in role-based access control. Mathematical
and Computer Modelling, 55(1):87–96.
Ma, X., Tian, Y., Zhao, L., and Li, R. (2013). Mining role
based on ranks. ICIC Express Letters. Part B, Ap-
plications: an Int. Journal of Research and Surveys,
4(2):319–326.
Mandala, S., Vukovic, M., Laredo, J., Ruan, Y., and Her-
nandez, M. (2012). Hybrid role mining for security
service solution. In Proc. of the 9th Int. Conf. on Ser-
vices Computing. IEEE.
Mitra, B., Sural, S., Atluri, V., and Vaidya, J. (2013). To-
ward mining of temporal roles. In Data and Ap-
plications Security and Privacy XXVII, volume 7964
of Lecture Notes in Computer Science, pages 65–80.
Springer.
Molloy, I., Chen, H., Li, T., Wang, Q., Li, N., Bertino, E.,
Calo, S., and Lobo, J. (2008). Mining roles with se-
mantic meanings. In Proc. of the 13th Symp. on Access
Control Models and Technologies. ACM.
Molloy, I., Chen, H., Li, T., Wang, Q., Li, N., Bertino, E.,
Calo, S., and Lobo, J. (2010). Mining roles with mul-
tiple objectives. In ACM Transactions on Information
and System Security. ACM.
Molloy, I., Park, Y., and Chari, S. (2012). Generative mod-
els for access control policies: Applications to role
mining over logs with attribution. In Proc. of the 17th
Symp. on Access Control Models and Technologies.
ACM.
Sandhu, R. S., Coyne, E. J., Feinstein, H. L., and Youman,
C. E. (1996). Role-based access control models. Com-
puter, 29(2):38–47.
Schlegelmilch, J. and Steffens, U. (2005). Role mining with
orca. In Proc. of the 10th Symp. on AccessControl
Models and Technologies. ACM.
SOX (2002). Sarbanes-oxley act of 2002, pl 107-204, 116
stat 745.
Takabi, H. and Joshi, J. B. (2010). Stateminer: An efficient
similarity-based approach for optimal mining of role
hierarchy. In Proc. of the 15th Symp. on Access Con-
trol Models and Technologies. ACM.
Uzun, E., Atluri, V., Lu, H., and Vaidya, J. (2011). An opti-
mization model for the extended role mining problem.
In Data and Applications Security and Privacy XXV,
pages 76–89. Springer.
Vaidya, J., Atluri, V., and Guo, Q. (2007). The role mining
problem: finding a minimal descriptive set of roles. In
Proc. of the 12th Symp. on Access Control models and
Technologies. ACM.
Vaidya, J., Atluri, V., and Guo, Q. (2010a). The role mining
problem: A formal perspective. ACM Transactions on
Information and System Security, 13(3):27.
Vaidya, J., Atluri, V., and Warner, J. (2006). Roleminer:
Mining roles using subset enumeration. In Proc. of the
13th ACM Conf. on Computer and Communications
Security. ACM.
Vaidya, J., Atluri, V., Warner, J., and Guo, Q. (2010b). Role
engineering via prioritized subset enumeration. IEEE
Transactions on Dependable and Secure Computing,
7(3):300–314.
Wang, J., Zeng, C., He, C., Hong, L., Zhou, L., Wong,
R. K., and Tian, J. (2012). Context-aware role min-
ing for mobile service recommendation. In Proc. of
the 27th Annual Symp. on Applied Computing. ACM.
Wong, R. K., Chu, V. W., Hao, T., and Wang, J. (2012).
Context-aware service recommendation for moving
AnalyzingQualityCriteriainRole-basedIdentityandAccessManagement
71
connected devices. In Int. Conf. on Connected Vehi-
cles and Expo.
Xu, Z. and Stoller, S. D. (2012). Algorithms for mining
meaningful roles.
Xu, Z. and Stoller, S. D. (2013a). Mining attribute-based
access control policies from rbac policies.
Xu, Z. and Stoller, S. D. (2013b). Mining parameterized
role-based policies.
Ye, W., Li, R., and Li, H. (2013). Role mining using boolean
matrix decomposition with hierarchy.
Zhang, D., Ramamohanarao, K., and Ebringer, T. (2007).
Role engineering using graph optimisation.
Zhang, D., Ramamohanarao, K., Ebringer, T., and Yann, T.
(2008). Permission set mining: Discovering practical
and useful roles.
Zhang, W., Chen, Y., Gunter, C., Liebovitz, D., and Malin,
B. (2013a). Evolving role definitions through permis-
sion invocation patterns.
Zhang, X., Han, W., Fang, Z., Yin, Y., and Mustafa, H.
(2013b). Role mining algorithm evaluation and im-
provement in large volume android applications.
Zhu, H. and Zhou, M. (2008). Roles in information sys-
tems: A survey. IEEE Transactions on Systems, Man,
and Cybernetics, Part C: Applications and Reviews,
38(3):377–396.
ICISSP2015-1stInternationalConferenceonInformationSystemsSecurityandPrivacy
72