From Confidential kNN Queries to Confidential Content-based
Publish/Subscribe
Emanuel Onica
1
, Hugues Mercier
2
and Etienne Rivi
`
ere
3
1
Faculty of Computer Science, Alexandru Ioan Cuza University of Ias¸ i, Romania
2
Institute of Computer Science, University of Neuch
ˆ
atel, Switzerland
3
ICTEAM, UCLouvain, Belgium
Keywords:
Publish/Subscribe, Data Confidentiality, Security, kNN Queries, Information Dissemination.
Abstract:
Content-based publish/subscribe (pub/sub) is an effective paradigm for information dissemination in dis-
tributed systems. In brief, publishers generate feeds of information, and subscriber clients register their inter-
ests with a pub/sub service tasked with delivering the published data to interested subscribers. Modern pub/sub
services are often externalized to public clouds. This brings economic advantages that are unfortunately over-
shadowed by associated security risks, in particular related to the confidentiality of both the published data
as well as of the subscriptions. Guaranteeing confidentiality for content-based pub/sub in an efficient fashion
is an active research area. A promising direction is to leverage specific cryptographic solutions that permit
the execution of the pub/sub service over encrypted data. In this article we describe a simple and general
methodology to derive new mechanisms for pub/sub confidentiality out of another category of data protection
schemes: confidential kNN query mechanisms designed for encrypted databases. We exemplify this frame-
work with a concrete use case. We believe that this initial step will lead to more secure and efficient adaptations
of kNN solutions to the pub/sub domain.
1 INTRODUCTION
Publish/subscribe (pub/sub) is a paradigm for infor-
mation dissemination that is particularly fit for large-
scale distributed systems and service-oriented archi-
tectures. A common pub/sub service deployment
consists in an overlay of brokers tasked with match-
ing subscriptions registered by clients of the service
(subscribers) with publications emitted by providers
of information (publishers). Two major models ex-
ist for the pub/sub paradigm in respect to the struc-
turing of information included in publications and
subscriptions and their matching: topic-based and
content-based. In topic-based pub/sub each subscrip-
tion and publication is associated with a topic of in-
terest and the match is decided according to these top-
ics. In content-based pub/sub the publication includes
a header composed of key-value attributes represen-
tative for the publication content. Subscriptions are
formed as conjunctions of constraints used to match
some or all of these attributes. An often met exam-
ple in literature (Eugster et al., 2003; Yang, 2010)
for content-based pub/sub is of a stock market sce-
nario, where a publication corresponding to a stock
quote includes attributes such as (symbol=’AMD’,
value=27.85, variation=0.03), and a subscrip-
tion could be formed as (symbol=’AMD’ and value
< 28). In our work we focus on content-based
pub/sub, the more expressive model, where a sub-
scriber has more flexibility in indicating interests.
The applications of pub/sub services have a wide
range, from stock market transactions (Bernstein and
Newcomer, 2009) to management of electronic med-
ical records (Narus et al., 2018). The brokers tasked
with matching publications and subscriptions in such
services are often externalized to public cloud infras-
tructures, via virtual machine instances. This brings
economic benefits but also introduces security chal-
lenges. The chance that a malicious virtual machine
will be co-located with a victim instance on Amazon
EC2, Google Compute Engine and Microsoft Azure
cloud infrastructures were estimated as ranging from
30% to 100%, depending on the power of the at-
tacker and the number of victim instances (Varadara-
jan et al., 2015). Leaks of information on co-located
virtual machines (Ristenpart et al., 2009) have been
documented for more than 10 years and are still un-
der scrutiny for finding appropriate protection mea-
sures (Han et al., 2017).
Specific to pub/sub is the fact that subscriptions
must be stored by untrusted brokers in order to be
matched with publications. In many use cases as ref-
Onica, E., Mercier, H. and RiviÃ
´
lre, E.
From Confidential kNN Queries to Confidential Content-based Publish/Subscribe.
DOI: 10.5220/0007950506770682
In Proceedings of the 14th International Conference on Software Technologies (ICSOFT 2019), pages 677-682
ISBN: 978-989-758-379-7
Copyright
c
 2019 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
677

Figure 1: High-level overview of subscription leaks in a cloud deployed pub/sub service.
erenced before, both subscriptions and publications
may convey sensitive information about the users’ in-
terests. Figure 1 displays a high-level overview of
an attack scenario where a malicious investor spies
on the subscriptions of another investor, which could
gain him unfair advantage on the stock market. A sim-
ilar example could be drawn for the case of publica-
tions that include private medical records of patients.
It is, therefore, necessary to protect the confidentiality
of both publications and subscriptions.
Unfortunately, “classical” end-to-end encryption
(e.g., AES) does not allow brokers to perform the
matching operation. Latest technology advances in
mainstream processors, such as Intel SGX or AMD
SEV (Mofrad et al., 2018) offer a possible context for
performing the matching in trusted hardware (Pires
et al., 2016). However, the availability of these
mechanisms depends on the infrastructure offered by
the cloud providers. Also, such hardware protec-
tion comes with limitations on memory size and vul-
nerability to side-channel attacks as the recent Fore-
shadow exploit (Van Bulck et al., 2018). Designing
a solution purely based on particular cryptographic
primitives that permit the matching remains, there-
fore, a valid option for offering data protection. Sev-
eral attempts in this area exist (Onica et al., 2016), but
current lack of adoption in practice is a proof that this
is still an open research direction.
We observe that matching between encrypted sub-
scriptions and publications bears similarities with
querying mechanisms used in encrypted databases,
in particular for implementing k-Nearest-Neighbor
queries (kNN for short). This paper presents our ini-
tial work towards a general methodology for trans-
forming kNN encrypted query schemes to solutions
for confidentiality-preserving pub/sub.
We start in Section 2 by presenting a generic
methodology for adapting any confidentiality-
preserving kNN scheme to the pub/sub context. We
follow in Section 3 with a concrete example of an
adaptation that complies with the defined generic
methodology. We discuss aspects related to security
and performance in Section 4, and conclude in
Section 5.
2 FROM KNN TO
CONTENT-BASED PUB/SUB
We first present the principles of kNN queries and
their similarities to pub/sub. We then identify the nec-
essary steps for adapting a plaintext kNN query over
a database to a plaintext pub/sub context. We finally
discuss the integration of encryption in the defined
methodology.
2.1 Similarities of kNN and Pub/Sub
In a database, a kNN query returns the k nearest
neighbors to a query point, according to some dis-
tance function (e.g., the Euclidean distance). There-
fore, the query is represented as a set of values over a
set of attributes (dimensions) and the reply is a set of k
entries from the database using the same schema. An
encryption scheme for kNN queries must both pre-
serve the confidentiality of the database records and
of the query point while still allowing to execute the
query.
In a content-based pub/sub system, subscription
constraints are expressed as ranges over some or all
ICSOFT 2019 - 14th International Conference on Software Technologies
678

of the attributes (dimensions), also respecting a pre-
defined publication schema. Subscriptions are stored
permanently at the brokers. Publications are points
in the attribute space, i.e., they define values for all
attributes. The result of the matching operation is a
decision on whether or not all the range constraints
of a subscription are matched by the values of the
publication. An encryption scheme for content-based
pub/sub has similar objectives as kNN queries: both
subscriptions and publications must be encrypted to
preserve their confidentiality, while untrusted brokers
must remain able to perform the matching.
2.2 From kNN to Pub/Sub in Plaintext
We consider a generic kNN query on a database,
which searches for the nearest k records A
1
, ..., A
k
to a
query Q, using the Euclidean distance. Both database
records and the query can be modeled as points in
a n-dimensional space, where each dimension corre-
sponds to a field in the database schema. Subscrip-
tions and publications in a pub/sub system can also be
modeled as points in a n-dimensional space. The main
difference comes from the fact that a subscription is
typically formed as a conjunction of constraints (e.g.,
<, >, =) over the values of individual dimensions.
A kNN query is typically evaluated by comparing
the Euclidean distances between the query point and
each of the database record points (e.g., dist(Q,A
1
)
compared with dist(Q, A
2
)), with the purpose of de-
termining which points are the closest. Matching in
pub/sub is fundamentally different. Pub/sub matching
requires to determine the exact relation (>, <, =) be-
tween each individual dimension in the subscription
point and the corresponding dimension in the publi-
cation point. This relation between individual dimen-
sions is not important in a kNN query, while the dis-
tance matters. It is the opposite for pub/sub. For in-
stance, in Figure 2, points A
1
= (1, 2) and A
2
= (3, 0)
are equally close to query point Q = (2, 1). A kNN
query for point Q will not distinguish between points
A
1
and A
2
. Therefore, in general, the result of a kNN
query does not help us determine the exact relation
between individual dimensions (x and y coordinates)
in Q and the corresponding values in either A
1
or A
2
.
However, there is a particular situation where the
kNN query result can be used to determine the rela-
tion between individual dimensions, which we lever-
age for content-based pub/sub. Let us consider a di-
mension of interest d
i
for which we want to find the
relation (>, <, =) between the corresponding values
in the query point and another point. We assume a
query point Q and two record points A
1
and A
2
where
all corresponding dimensions in A
1
and A
2
are equal,
except for the dimension of interest d
i
. Furthermore,
let us consider the point A
m
at the middle of the seg-
ment [A
1
A
2
], and assume that we know the relation
(<, >, =) between the value of d
i
in points A
1
and
A
2
. Figure 3 gives an example of this context for
bi-dimensional points A
1
= (1, 1), A
2
= (5, 1), A
m
=
(3, 1) and Q = (2, 2), where we consider d
1
as the
dimension of interest (d
1
corresponds to the x axis).
We want to determine the exact relation between the
value of d
1
in point Q and the value of d
1
in the point
A
m
using Q as kNN query over points A
1
and A
2
. In
this example the kNN query result for query point Q
would clearly show that dist(Q, A
1
) < dist(Q, A
2
).
Let Q
P
be the projection of query Q on the
axis determined by A
1
and A
2
. It is trivial to
prove, under the assumptions of the above context,
that if dist(Q, A
1
) < dist(Q, A
2
) ⇒ dist(Q
P
, A
1
) <
dist(Q
P
, A
2
). More generally, if dist(Q, A
1
) ∼
dist(Q, A
2
) ⇒ dist(Q
P
, A
1
) ∼ dist(Q
P
, A
2
) where
∼ ∈ {<, >, =}. Since A
1
, A
2
, A
m
and Q
P
are on
the same axis and have all dimensions equal except of
d
1
, from dist(Q
P
, A
1
) < dist(Q
P
, A
2
) follows that d
1
in Q
P
is smaller than d
1
in A
m
. In the assumed con-
text where Q
P
is the projection of Q on an axis where
all dimensions are equal except d
1
, we also clearly
can infer that d
1
in Q is equal to the value of d
1
in
Q
P
. Therefore, the relation (<, >, =) we determined
between d
1
in Q
P
and d
1
in A
m
is always preserved
when comparing d
1
in Q with d
1
in A
m
. This leads to
our searched result: d
1
in Q is smaller than d
1
in A
m
.
2.3 From Encrypted kNN to Encrypted
Pub/Sub
We can now define a generic approach for adapting
any confidential kNN query to a confidential pub/sub
solution. In the pub/sub scenario a subscription S will
be represented using the format of records A
1
and A
2
considered in the example above, and a publication
P will take the representation of the kNN query Q.
More precisely, for each dimension of interest d
i
in
a subscription, a subscriber will consider two points
S
i1
and S
i2
, such that d
i
is the middle of the segment
[S
i1
S
i2
]. All other dimensions in S
i1
and S
i2
can be
chosen randomly, but must be the same for the two
points. Note that this does not restrict the subscriber:
any value can be used for any dimension of interest
d
i
. A publication P will simply resemble the query Q.
Cryptographic schemes for confidentiality-
preserving kNN queries typically define a type of
encryption for both query and records, such that
the query result can be obtained over the encrypted
form without leaking information about the values in
the dimensions. We distinguish two variants in the
From Confidential kNN Queries to Confidential Content-based Publish/Subscribe
679

Figure 2: Distance comparison between points in a kNN
query.
Figure 3: Distance comparison between points in the
pub/sub case.
operation of the confidential kNN query:
1. The scheme provides or exposes the ordering of
the k records based on their distance from the
query;
2. The scheme only provides the final result of the k
closest records to the query point, without leaking
anything else.
In the first variant it is enough to represent sub-
scriptions as records, as defined above, encrypt them,
proceed similarly with encrypting publications in the
same way the queries are encrypted, and execute the
confidential kNN query with P over all points in pairs
(S
i1
, S
i2
) up to when we can extract the comparison
result between the distances. Then, from this result:
dist(S
i1
, P) ∼ dist(S
i2
, P) (where ∼ ∈ {<, >, =}),
following the reasoning in Section 2.2, we obtain the
relation between the value of d
i
in P and of d
i
in S,
which allows us to decide on the pub/sub match for
dimension d
i
. This is similar to our example, where
we obtained the relation between dimension d
1
in Q
and d
1
in A
m
.
In the second variant, we can execute a 1NN con-
fidential query with the encrypted P over each pair
of encrypted points (S
i1
, S
i2
) sequentially. This will
obviously show which of S
i1
or S
i2
is closer to P and
implicitly provide the needed distance comparison re-
sult dist(S
i1
, P) ∼ dist(S
i2
, P) for proceeding as above
to determine the pub/sub match.
We note that the second variant can be applied
in any situation. Defining the first variant is merely
for efficiency purposes: if the ordering is exposed,
i.e., the scheme mechanism permits comparing dis-
tances between encrypted points, we can leverage
it. Furthermore, the structure of the encrypted sub-
scriptions and how they are stored might vary (e.g.,
if the scheme permits evaluating subscription cover-
age, encrypted subscriptions could be organized in
containment trees (Barazzutti et al., 2017)). Execut-
ing 1NN queries sequentially over each pair of en-
crypted points might, therefore, require reorganizing
their storage. In such a case, if a scheme exposes the
ordering in its default query run, that scheme might
be more efficient to use than a scheme that does not.
3 APPLICATION EXAMPLE
The first and only application of a confidentiality
preserving kNN query to pub/sub was presented
by (Choi et al., 2010), using asymmetric scalar-
product-preserving encryption (ASPE), a scheme in-
troduced earlier by (Wong et al., 2009). The adapta-
tion follows a path similar to the methodology defined
in Section 2, but considers mostly the case of one sin-
gle dimension/attribute for publications and subscrip-
tions schemas. Also, it does not include any general-
ization or analysis of the adaptation procedure. We
addressed the multidimensional case and hardened
the security of the original scheme, offering proofs
of security in (Onica et al., 2015). However, the main
focus of our previous work was on performing key
updates in a secure pub/sub context, and a general-
ization that could be applied for adapting other cryp-
tographic schemes was not defined. Nevertheless,
ASPE can serve as a valid proof for the generic adap-
tation methodology we are proposing, and we sum-
marize it in the following.
For each dimension d
i
in a subscription, the sub-
scriber prepares two points S
i1
and S
i2
as described
in Section 2. These points are encrypted as: S
0
i1
=
M
T
(S
i1
, −0.5kS
i1
k)
T
and S
0
i2
= M
T
(S
i2
, −0.5kS
i2
k)
T
where kS
i1
k and kS
i2
k represent the Euclidean norm
of the two points, and M is an invertible non-
orthogonal matrix representing the encryption key.
This is the same method of encryption used for
database records in the original confidential kNN
scheme (Wong et al., 2009). The publisher encrypts
the publication as: P
0
= M
−1
q(P, 1)
T
, where q is a
random positive obfuscation factor. Again, this is
the same method of encryption used for the kNN
query point (Wong et al., 2009). Finally, an un-
ICSOFT 2019 - 14th International Conference on Software Technologies
680

trusted broker evaluates the result of (S
0
i2
− S
0
i1
)P
0
=
q0.5(dist(S
i1
, P)
2
− dist(S
i2
, P)
2
). Simply comparing
this result with 0 exposes the relation dist(S
i1
, P) ∼
dist(S
i2
, P) (where ∼ ∈ {<, >, =}), which places the
scheme in the first variant identified in Section 2, and
enables the decision on pub/sub matching.
As described earlier, the construction of the en-
cryption mechanism is orthogonal to the steps in our
methodology for matching encrypted subscriptions
against encrypted publications. The adaptation just
requires the particular representation of the subscrip-
tion points before the encryption, and determining the
distance comparison result following the execution of
the query scheme.
4 DISCUSSION
Although our adaptation methodology could theoret-
ically apply to any confidentiality preserving kNN
query scheme, several aspects related to security and
performance deserve a more thorough discussion. We
focus on these details in the following.
4.1 Security Aspects
First of all, the security of the obtained confidential
pub/sub solution simply relies on the fact that the
original encryption scheme is secure. Our adapta-
tion does not influence the encryption scheme in any
manner. Essentially, the only particularity is that it
considers two specific subscription points for each di-
mension of interest, but these subscription points are
first generated in plaintext and encrypted as any other
record point in the original scheme. As long as the
original encryption scheme is proven secure this guar-
antee holds for any encrypted point. The property that
allows adapting a confidential kNN query scheme for
pub/sub matching is preserving the distance compar-
ison results in the encrypted form with respect to the
plaintext context. As determined from variants 1 and
2 in Section 2, this preservation is normally valid. We
note that preserving the distance comparison results is
a weaker property than preserving the actual distance
values after encryption. The latter is not desirable, re-
sulting in confidentiality leaks, as discussed by (Wong
et al., 2009).
4.2 Performance Aspects
Evaluating the pub/sub matching can increase up
to a quadratic complexity compared to the plaintext
matching, due to the subscription points representa-
tion. For each dimension of interest d
i
in an orig-
inal subscription, the two corresponding points that
are created S
i1
and S
i2
as required by our adaptation,
must include as the other random chosen dimensions
the complete set of attributes in a publication. If we
consider n such fields, this might lead to n operations
per attribute of interest in the evaluation of the match-
ing, and consequently to n
2
operations if we have a
complete subscription. This is also multiplied with
factor 2 per subscription, since for each attribute we
have two corresponding points that are created. How-
ever, in most practical settings, especially for a high
number n of attributes, it is unlikely that most sub-
scriptions will set constraints on each field of a pub-
lication. Also, this potential increase in complexity
is highly dependent on the actual encryption scheme
and the means it uses to provide the distance compar-
ison result, which can be subject of various optimiza-
tions. For instance, in the example case of ASPE,
the two corresponding points created for each field
of a subscription can be aggregated into one point in
the encryption phase, therefore removing the factor
2 above. We observed this in previous work (Onica
et al., 2015), and noted that it is actually desirable for
stronger security reasons. Also, a scheme might also
permit evaluating if an encrypted subscription cov-
ers another (i.e., the case where a publication match-
ing the covering subscription will match all covered
subscriptions), which can lead to consistent improve-
ment, by minimizing the number of evaluated sub-
scriptions (Barazzutti et al., 2017). However, cov-
erage support can also lead to leaks of information
on the subscription domain (Raiciu and Rosenblum,
2006).
5 CONCLUSION
We presented our current work in devising a generic
methodology for deriving confidentiality preserving
pub/sub schemes based on solutions for confiden-
tiality preservation in kNN queries. We explained
the reasoning behind the methodology and the series
of simple steps needed for such an adaptation. We
also exemplified how these were already applied for
ASPE, a scheme initially designed for kNN queries
and later transposed to pub/sub. The main advan-
tage of our proposed methodology is that it is or-
thogonal to the specificities of the encryption scheme.
Therefore, we believe that it could be applied to other
schemes dedicated for preserving kNN confidential-
ity. In particular there are many solutions derived
from ASPE that could potentially be adapted to the
pub/sub realm (Tzouramanis, 2017; Tzouramanis and
Manolopoulos, 2018; Zhu et al., 2016; Zhu et al.,
From Confidential kNN Queries to Confidential Content-based Publish/Subscribe
681

2013), as well as completely different solutions such
as RASP (Xu et al., 2014). Our future work is to
adapt some of these schemes and observe their per-
formance versus the security they provide, using the
proposed methodology. We believe that our work can
be an important first step for positive outcomes in the
research and exploration of such new solutions for
pub/sub confidentiality.
REFERENCES
Barazzutti, R., Felber, P., Mercier, H., Onica, E., and
Riviere, E. (2017). Efficient and confidentiality-
preserving content-based publish/subscribe with pre-
filtering. IEEE Transactions on Dependable and Se-
cure Computing, 14(3).
Bernstein, P. A. and Newcomer, E. (2009). Principles of
transaction processing. Elsevier/Morgan Kaufmann
Publishers, Amsterdam, Netherlands; Boston, USA.
Choi, S., Ghinita, G., and Bertino, E. (2010). A privacy-
enhancing content-based publish/subscribe system us-
ing scalar product preserving transformations. In Pro-
ceedings of the 21st Database and Expert Systems Ap-
plications, DEXA 2010.
Eugster, P. T., Felber, P., Guerraoui, R., and Kermarrec, A.-
M. (2003). The many faces of publish/subscribe. ACM
Computing Surveys, 35(2).
Han, Y., Chan, J., Alpcan, T., and Leckie, C. (2017). Using
virtual machine allocation policies to defend against
co-resident attacks in cloud computing. IEEE Trans-
actions on Dependable and Secure Computing, 14(1).
Mofrad, S., Zhang, F., Lu, S., and Shi, W. (2018). A com-
parison study of Intel SGX and AMD memory encryp-
tion technology. In Proceedings of the 7th Interna-
tional Workshop on Hardware and Architectural Sup-
port for Security and Privacy, HASP 2018.
Narus, S. P., Rahman, N., Mann, D. K., He, S., and
Haug, P. J. (2018). Enhancing a commercial EMR
with an open, standards-based publish-subscribe in-
frastructure. AMIA Annual Symposium Proceedings,
2018.
Onica, E., Felber, P., Mercier, H., and Rivi
`
ere, E.
(2015). Efficient key updates through subscription re-
encryption for privacy-preserving publish/subscribe.
In Proceedings of the 16th ACM/IFIP Annual Middle-
ware Conference, Middleware 2015.
Onica, E., Felber, P., Mercier, H., and Rivi
`
ere, E. (2016).
Confidentiality-preserving publish/subscribe: A sur-
vey. ACM Computing Surveys, 49(2).
Pires, R., Pasin, M., Felber, P., and Fetzer, C. (2016). Secure
content-based routing using Intel software guard ex-
tensions. In Proceedings of the 17th ACM/IFIP Inter-
national Middleware Conference, Middleware 2016.
Raiciu, C. and Rosenblum, D. S. (2006). Enabling confiden-
tiality in content-based publish/subscribe infrastruc-
tures. In Proceedings of the Second IEEE/CreatNet
International Conference on Security and Privacy in
Communication Networks, Securecomm 2006.
Ristenpart, T., Tromer, E., Shacham, H., and Savage, S.
(2009). Hey, you, get off of my cloud: exploring infor-
mation leakage in third-party compute clouds. In Pro-
ceedings of the 16th ACM Conference on Computer
and Communications Security, CCS 2009.
Tzouramanis, T. (2017). Secure range query processing
over untrustworthy cloud services. In Proceedings of
the 21st International Database Engineering & Appli-
cations Symposium, IDEAS 2017.
Tzouramanis, T. and Manolopoulos, Y. (2018). Secure
reverse k-nearest neighbours search over encrypted
multi-dimensional databases. In Proceedings of the
22nd International Database Engineering & Applica-
tions Symposium, IDEAS 2018.
Van Bulck, J., Minkin, M., Weisse, O., Genkin, D., Kasikci,
B., Piessens, F., Silberstein, M., Wenisch, T. F.,
Yarom, Y., and Strackx, R. (2018). Foreshadow: Ex-
tracting the keys to the Intel SGX kingdom with tran-
sient out-of-order execution. In Proceedings of the
27th USENIX Security Symposium.
Varadarajan, V., Zhang, Y., Ristenpart, T., and Swift, M.
(2015). A placement vulnerability study in multi-
tenant public clouds. In Proceedings of the 24th
USENIX Security Symposium.
Wong, W. K., Cheung, D. W.-l., Kao, B., and Mamoulis,
N. (2009). Secure kNN computation on encrypted
databases. In Proceedings of the 35th ACM SIGMOD
International Conference on Management of Data,
SIGMOD 2009.
Xu, H., Guo, S., and Chen, K. (2014). Building confidential
and efficient query services in the cloud with RASP
data perturbation. IEEE Transactions on Knowledge
and Data Engineering, 26(2).
Yang, L. T. (2010). Research in Mobile Intelligence: Mo-
bile Computing and Computational Intelligence. John
Wiley & Sons, Hoboken, USA.
Zhu, Y., Huang, Z., and Takagi, T. (2016). Secure and
controllable kNN query over encrypted cloud data
with key confidentiality. Journal of Parallel and Dis-
tributed Computing, 89.
Zhu, Y., Xu, R., and Takagi, T. (2013). Secure kNN query
on encrypted cloud database without key-sharing. In-
ternational Journal of Electronic Security and Digital
Forensics, 5(3/4).
ICSOFT 2019 - 14th International Conference on Software Technologies
682