International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 1 (2016)pp 738-743
© Research India Publications. http://www.ripublication.com
An efficient Multi-Keyword Synonym Ranked Query over Encrypted
Cloud Data using BMS Tree
Veerraju Gampala,
Department of Computer Science and Engineering,
GMR Institute of Technology, Rajam, Andhra Pradesh, India.
E-mail: [email protected]
Sreelatha Malempati,
Dept. of Computer Science and Engineering,
R.V.R & J.C. College of Engineering,
Chowdavaram, Guntur, Andhra Pradesh, India.
E-mail: [email protected]
database vendors such as Oracle and IBM with IBM
BlueMix are adding cloud sustenance to their databases.
To protect data privacy, confidentiality, and data security,
sensitive data like personal health records, e-mails, tax
documents, photo albums, financial transactions, and so on,
have to be encrypted by data owners before outsourcing to
the public cloud [16]. However, the traditional plaintext
keyword search data utilization service is obsolete.
Downloading all the data and decrypting at the data user side
is trivially impractical duo to huge amount of bandwidth cost
is required in cloud scale systems. The data can be easily
searched and utilized otherwise no purpose of storing data in
the cloud. Thus, exploring effective and privacy preserving
search over encrypted cloud data is of most prominence. This
is a very challenging issue; it degrades performance of
system usability and scalability. It is very difficult to meet
the requirements of system usability, performance, and
scalability, by considering the large number of on-demand
data users and large number of outsourced data documents in
the cloud. To meet effective data retrieval, the huge amount
of documents demands the cloud server to perform relevance
ranking as a result, instead returning all result documents.
Such ranking system facilitates data users to find the most
relevant data rapidly, rather than burdensome sorting through
every match in the content collection [17]. Ranked search
eliminates the unnecessary network traffic by sending only
most relevant documents, which is very much desirable in
“pay-as-you-use” cloud model. Such ranked systems should
not leak any keyword related information to the cloud server
for privacy protection. Such ranking systems should support
fuzzy keyword and multi-keyword search, as a single
keyword search often gives far too harsh results.
In general, the data users enters set of keywords in web
search engines to retrieve most relevant data instead of only a
single keyword to indicate interest of search. “Coordinate
matching” [18] is widely used in plaintext information
retrieval systems. It is an efficient similarity measure as
many matches as possible to refine the result relevance.
However, how to apply this technique on encrypted cloud
data is still a challenging task because to meet privacy
preserving requirements like data privacy, user privacy,
index privacy, query privacy, keyword privacy, audit, access
control, integrity, authenticity, confidentiality and
unlinkability. In this paper, we propose a method to solve the
problem of Multi-keyword Synonym Ranked Query over
encrypted cloud data (MSRQE), while preserving strict
system wise privacy in cloud computing paradigm. The main
Abstract
Cloud computing is embryonic technology as a new
computing model in several business domains. Large
numbers of large-scale organizations are startingto shift the
information on to the cloud environment. However, shifting
the data to the cloud relieves the organizations from the
monotonous tasks of organization management, minimizes
maintenance
costs
and
also
less
hands
on
management.However, security and privacy become key
security issues when data owners outsource their private data
onto untrusted public cloud servers. However, traditional
keyword plain text search is obsolete. Several techniques
have been developed to preserve the privacy of sensitive data
in the cloud environment. But existing searching techniques
over encrypted cloud data considers only exact or fuzzy
keyword or multi-keyword, but not synonym based ranking
searching including multi-keyword. In this paper, we propose
an efficient multi-keyword synonym query over encrypted
cloud data by retrieving top k scored documents.The vector
space model and TFIDF model are used to construct index
and query generation. The KNN algorithm used to encrypt
index and query vectors. We construct a special tree called
Balanced M-way Search (BMS) Tree for indexing and
propose a Depth First Search Technique (DFST) algorithm to
achieve efficient multi-keyword synonym ranked search. The
efficiency and accuracy of DFST algorithm are illustrated
with an example, BMS tree, it takes sub-linear time
complexity.
Keywords:privacy preserving, multi-keyword search,
synonym keyword search, BMS tree, DFST algorithm.
INTRODUCTION
Cloud computing is the broad fantasized dream of computing
as a service, where data owners of cloud can remotely store
their sensitive and private data like emails, health records and
personal photos onto the cloud. With the quality applications,
we can get services on-demand from a shared environment
with computing resources from anywhere in the world. It is a
great tractability and cost saving to motivating both industry
and individual to outsource their local private complex data
management system onto the cloud. The state-of-the-art
cloud-computing paradigm has generated a substantial
interest in both industry and academia, resulting in a number
of famous marketable and individual cloud computing
services provided by several cloud service providers,
examples from Amazon web services, Google App engine,
Microsoft Azure, Aneka, Yahoo, and Sales force. Top
738
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 1 (2016)pp 738-743
© Research India Publications. http://www.ripublication.com
contributions are four aspects: the first one multi-keyword
search, the second synonym based search, third similarity
ranked search and the last is efficient data retrieval with
BMS tree and DFST algorithm. Our example analysis further
shows efficient and accurate top k documents retrieval of the
proposed scheme.
measure coordinate matching while meeting different privacy
requirements in two different threat models. They enhanced
ranked search mechanism to support dynamic data
operations. In no above techniques supported semantic
search and synonym-based search.
In our work, we designed an efficient, secure multi-keyword
synonym ranked searching technique over encrypted cloud
data by designing BMS tree and DFST algorithm
RELATED WORK
Many searching techniques over encrypted cloud data have
proposed. S. Deshpande [7] suggested a technique searching
over encrypted cloud data using fuzzy keywords. They used
Edit distance to quantify keyword similarity and developed
two techniques on constructing fuzzy keyword sets to
achieve optimized storage and representation overheads.
Cong wang et al. [1] Has proposed a method ranked keyword
search over encrypted cloud data using keyword frequency
and order preserving encryption. It supports only single
keywords at a time. Is the keyword frequency deciding
document file score. Rank given to every file based on the
relevance score of that file. Top ranked files have sent to
users instead all files. To enrich search functionality N. Cao
et al. [2] Have proposed a scheme supporting conjunctive
keywords search. It is privacy – preserving multi-keyword
ranked search technique using symmetric encryption. M.
Chou et al. [6] proposed a solution for fuzzy multi-keyword
search over encrypted cloud data using privacy aware Bed
Tree.They used a co-occurrence probability approach to
identify useful multi-keywords for publishingdata,
documents and relevant fuzzy keyword sets constructed
using edit distance. They constructed index tree for all data,
documents, where each leaf node having the hash value of a
keyword, one or two data vectors that represents n- gram of
that keyword and bloom filters for each edit distance value.
C. Wang et al. [3] Suggested a technique to build storage
efficient similarity keyword set with a given document
collection, edit distance as a similaritymetric. Based on that,
they built a private trie traverse-searching the index to
achieve similarity search functionality with constant time
complexity. C.Yang et al. [4] Designed a scheme adopting
three sparse matrices instead dense matrix pair in MRSE to
encrypt index, they combined their scheme with a bloom
filter to gain the ability for index updating. D.X.Song. D et
al. [8] Proposed a method, remotely searching over encrypted
data using an untrusted server and provided proof for the
resulting crypto system. They supported controlled, hidden
search and query isolation. C. Wang et al. [10] Proposed an
efficient Ranked Searchable Symmetric Encryption scheme
(RSSE). It supports ranked keyword search, security
guarantee and efficiency with minimum communication cost.
B. Wang et al. [11] suggesteda novel multi-keyword fuzzy
search scheme by exploiting the locality sensitive hashing
techniques. They achieved fuzzy matching through
algorithmic design rather than expanding index file and it
eliminates the need of predefined dictionary. N. Cao et al.
[12] proposed two Multi - keyword Ranked Search over
Encrypted cloud data (MRSE) schemes based on a similarity
SYSTEM MODEL
In this paper, we considered a cloud computing system
model involves three different entities. Those are Data
Owner, Cloud Service Provider and Data user as illustrated
in Fig. 1. The responsibility of each entity is as follows:
Data Owner (DO): DO has a collection data documents DC=
{d1, d2…, dm} with sensitive information to be outsourced to
the cloud server. To provide data privacy, the documents are
encrypted before outsourcing. DO creates a dictionary based
on keywords extracted from the all m documents based on
Term Frequency Inverted Document Frequency (TFIDF)
[13] which is described in section 4. The dictionary includes
synonyms of each keyword from the thesaurus [14]. The
dictionaryishavingandkeywords, and for each keyword may
have t synonyms, so that the dictionary size is n × t. DO
creates an index vector for each document based on the
keywords extracted from the document. The size of the index
vector is equal to the number of keywords in the dictionary
that is an. Each dimension in the index vector stores sum of
the frequency of keyword and corresponding synonyms in
the dictionary is denoted as term frequency (TF) in our
system. Index vectors of all documents are encrypted before
outsource to the cloud. DO create query vector based on
keywords entered by Data user. To provide user privacy,
query vector encrypted, as Trapdoor and send to Data user.
The data owner sends search access control to the authorized
data user.
Data users: Data users are the users who accessing sensitive
data from the cloud. The cloud server searches keywords or
synonyms related to documents, which are interested to data
user and sends to the data owner. The data user receives
trapdoor and searches access control of data owner and sends
trapdoor and access control to the cloud server to retrieve
required documents from the cloud.
Cloud Service Provider (CSP): Cloud server receives
encrypted documents and encrypted index vectors from data
owner and stores into data owner’s cloud storage. Cloud
server having the capability to take the data request from user
and check the search access control of the user. It will
retrieve the documents from cloud storage depending upon
the privileges to access number of documents. To increase
the document retrieval accuracy from cloud server, the top
scored (ranked) documents return to data user from the cloud
server. Fig. 1 shows the architecture of multi-keyword
synonym query over encrypted cloud data.
739
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 1 (2016)pp 738-743
© Research India Publications. http://www.ripublication.com
Figure 1: Architecture of multi-keyword synonym query over encrypted cloud data
Kn1, kn2,……,Knt}
Sample Dictionary used in our model is shown in Table2.
W: the subset of keywords and synonyms of D in search
query.
F: set of index vectors associated with DC denoted by F=
{F1, F2,….., Fm}.
Fu: index vector stored in the index tree at node u
Iu: encrypted form Fu
Ŧ: unencrypted BMS tree for all documents
Ĭ: searchable encrypted index tree generated from Ŧ
I: set of encrypted index vectors for F denoted by I = {I1,
I2,…., Im}.
TF: The sum of keyword frequency and corresponding
synonyms (in the dictionary) frequency in the document. In
general, Term frequency means the number of times the term
appears in the document.
Q: query vector includes the keywords or synonyms
interested to search by data users
T: Trapdoor for a set of keywords in W based on query
vector Q. It is encrypted form of query vector.
RankedList Ek: top ranked documents list according to their
relevance to query vector. It is a descending order list based
on the documents scores.
Threat model:
The cloud server is measured as “honest-but-curious” [15] in
our proposed scheme. The cloud server follows the proposed
method specification and also examines data in its cloud
storage and data which are received from data user during the
processing to learn additional information. We consider one
threat model for our system with different attack capabilities
that is as follows:
Known ciphertext model: In this model, the cloud server
knows only encrypted documents and encrypted index
vectors, which are outsourced from data owner.
PRELIMINARIES
Notations:
DC: is a collection of m plaintext document files, as DC =
{d1, d2…, dm}.
ED: is a collection of encrypted documents stored in cloud
storage expressed as ED = {c1, c2,..., cm}.
D: it is a keyword and synonym dictionary including n
keywords and for each keyword t synonyms. It is expressed
as Dn×t= {k11, k12,……., k1t
K21,……….., k2t
..
..
Table 1: D7×5 Dictionary with keywords and corresponding synonym
D[i][j]
1
2
3
4
5
6
7
keyword
1
Privacy
Rank
Document
Frequency
Term
Trapdoor
Information
Synonym1
2
Secrecy
Rampant
File
Occurrence
Word
Access
Material
Synonym2
3
Confidentiality
Fecund
Text
Rate
Period
Entrance
Data
740
Synonym3
4
Solitude
Overgrown
Article
Regularity
Tenure
Doorway
Info
Synonym4
5
Isolation
Vigorous
Essay
Incidence
Stint
Hatch
Evidence
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 1 (2016)pp 738-743
© Research India Publications. http://www.ripublication.com
the relevance between the query and the corresponding
document.
TFt,dweight (wt,d) = log (TFt,d)
IDFtweight (wt,q)= log (N/DFt)
𝑤𝑡,𝑑
Normalized TFt,d weight =
2
The Vector Space Model for Relevance Score:
The representation of a pool of documents as vectors in a
common vector space is called vector space model [13]. This
model along with TF-IDF is used in plain text information
retrieval, which supports ranked multi-keyword search
efficiently. Here, TFt,d is the term frequency defined as the
number of times keyword (term) t appears in the document d.
The number of documents that contains keyword t is known
as document frequency (DFt). Inverse Document Frequency
(IDFt) of a keyword t obtained fromthe total number of
documents is divided by document frequency. That means,
TF denotes the importance of keyword in the document file
and IDF specifies the degree of distinction in the whole
document file collection. Every document file is denoted
with a vector, whose elements are the normalized TF weights
of keywords in this document file. The query is also denoted
with a vector, whose elements are normalized IDF weights of
query keywords in the document collection. Naturally, the
size of the query vector and document vector are equal to the
total number of keywords in the dictionary. The dot product
between query vector and document vector gives the
relevance score of the corresponding document; it quantifies
Normalized IDFt weight =
√∑ 𝑁
𝑡=1(𝑤𝑡,𝑑 )
𝑤𝑡,𝑞
2
√∑𝑁
𝑡=1(𝑤𝑡,𝑞 )
Relevance score of query vector Q and document vector or
index vector Fu of node uis calculated as follows:
Score (Fu, Q) = Fu . Q = ∑𝑡∈𝑊𝑞 𝑇𝐹𝑢, 𝑡 × 𝐼𝐷𝐹𝑡 Where TFu,t is
the TF value of term t at node u
(1)
MSRQE DESIGN
In this section, we first construct BMS tree based on the
vector space model and describe DFST searching algorithm
for unencrypted BMS tree. Later, we described a secure
multi-keyword synonym ranked query overthe encrypted
BMS tree index.
Figure 2: An example of BMS tree and search process
D[i] = max {ui.F[j] for i=1 to 5} for each j=1 to n;
BMS Tree Index Construction:
In the process index tree construction, we generate node for
each document in the document collection. These nodes are
act as leaf nodes in the tree. Then, the internal nodes are
formed based on these leaf nodes. The index tree
construction process is described in the algorithm 1. An
example of BMS index tree for our scheme is shown in Fig.
2 which is constructed on plaintext. The data structure of the
node is defined as (ID, F, child [], DID), where ID is a
unique id generated using GenID() function, F is index
vector, child[] is pointers to children of the node and DID is
a document ID. In the algorithm, we used two variables
CurrentNodeCollection and TempNodeCollection to store
collection of nodes. CurrentNodeCollection stores the set of
currently processing nodes which have no parents and
TempNodeCollection stores set of newly formed nodes. Fu[i]
always stores the biggest TF value of wiamong its children.
The possible largest relevance score of its children is
estimated using this technique.
Algorithm 1 BuildBMSIndexTree(DC)
For each data document Ddid in DC do
Construct leaf node l for Ddid l.ID=GenID(),
l.child[i]=null for i=1,…, b; l.DID=DID, and
F[i]=TFDdid,ki for i=1,…, n;
Insert l to CurrentNodeCollection;
End for
While the number of nodes in CurrentNodeCollection is
more than 1 do
For each five of nodes u1, u2, u3, u4, and u5 in
CurrentNodeCollection do
Generate a parent node u for u1, u2, u3, u4, and
u5with u.ID=GenID(), u.child[i] = uifor i = 1to 5;
u.DID = 0, and D[i] = max{ui.F[j] for i=1 to 5}
for each j=1 to n;
Insert u to TempNodeCollection;
End for
741
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 1 (2016)pp 738-743
© Research India Publications. http://www.ripublication.com
The remaining nodes (less than 5 nodes) in
CurrentNodeCollection generate a parent node u
like above;
Insert u to TempNodeCollection;
Replace
CurrentNodeCollection
with
TempNodeCollection
and
then
free
the
TempNodeCollection;
End while
Return only one node, left in the CurrentNodeCollection
called the root node;
calling BuildBMSIndexTree (DC) algorithm. Now split the
each index vector Fuof node u in to two random
vectorsFu1and Fu2 using S bit vector. If S[i] = 0 then Fu1[i]
and Fu2[i] will be set to same as Fu1[i]; if S[i]=1 then Fu1[i]
and Fu2[i] will be set to two random values whose sum equals
to Fu[i]. The encrypted index vector for the Index treeis Iu=
{M1T.Fu1, M2TFu2}. The similar process applies to all the
documents in the index treeto generate encrypted index
vectors.
Trapdoor (W):
The interested keywords entered by a data user of W, this
function generates a bit vector Q based on keywords or
synonyms in D, if keywords available in D then set the IDF
value of the term to a specific dimension in Q respectively
otherwise set 0. Now apply the same splitting technique
which we have applied above to Q, divide into Qand Qto
provide search privacy, encrypt query vector before
outsourcing. The encrypted query vector is
T = {M1-1 Q,M2-1 Q}.
Search Process using DFST:
The search process of MSRQE scheme is the recursive
function upon the BMS tree name as Depth First Search
Technique algorithm. We create a result documents as
RankedList, whose element is denoted as (Score, DID). Here,
the score is the relevance score between Fdid and query vector
Q, which is calculated using formula(1). The RankedList
stores top k scored documents to query. The elements of
RankedList are in descending order according to score
function during the search process. The DFST algorithm is
presented in algorithm 2. Kth score is a smallest relevance
score in RankedList.
Algorithm 2 DFST(IndexTreeNode u)
If the node u is not a leaf node then
If Score(Fu, Q) > kth score then
Sort the children of u in descending order according to
scores of children
For i=1 to the number of children of u do
GDFS(u.child[i]);
End for
Else
Return;
End if
Else
If Score(Fu, Q) > kth score then
Delete the element with a smallest relevance score from
RankedList;
Insert a new element (Score (Fu, Q), u.ID) and sort all
elements of RankedList in descending order;
End if
Return;
End if
Query (T, Iu):
The query is processed using the inner product of trapdoor T
and encrypted index vector Iuby cloud server, and then we
will get a score for the document which is stored in node u in
BMS tree. The same inner product is applied to all nodes in
index tree, so that it generates scores for leaf nodes and
internal nodes. The relevance score calculation of the
document is as follows:
Score=Iu .T
= {M1T.Fu1, M2TFu2} × {M1-1 Q,M2-1 Q}
= M1TFu1M1-1QM2TFu2M2-1Q
Fu1M1 × M1-1. QFu2M2 × M2-1.Q
Fu1QFu2Q
Fu TQ
Illustrations of Proposed scheme:
An example of the balanced m-way search tree-based index
with the document collection DC= {di| i=1 to 15} and
dictionary size n=4. In the construction process, first leaf
nodes are created from documents. Then, internal nodes are
formed based on leaf nodes. The Fig. 2 shows the example,
BMS tree and also shows an example of the search process,
in which the query vector Q= (0.91, 0, 0.8, 0.45). In this
example the rank k=4, it means the cloud server returns top
four scored documents to the data user. According to DFST
search algorithm, the search starts at the root node and
reaches d15 leaf node through l13 internal node. The relevance
score of d15to query vector is 8.43. After that, the leaf nodes
d12, d13, and d14 reached in sequential order with relevance
scores 5.33, 2.61, and 2.61 respectively. All the four
documents scores and did are stored in RankedList in
descending order, i.e. d15, d12, d13, and d14. After that, the leaf
node d11 reached with relevance score 2.15, but this score is
less than the kth score in RankedList. Next, the algorithm
searches the subtree through l12 and finds the relevance score
of l12 to query vector is 7.63 which is greater than kth score in
RankedList so search process reaches d7. The relevance score
of d7 to query vector is 6.72 so, d14 is replaced with d7 and
then sorted the RankedList in descending order. Next, d8 is
reached with relevance score 6.03 so, d13 is replaced with d8.
MSRQE scheme:
We construct a basic MSRQE scheme based on a secure
KNN algorithm [5]. The MSRQE is designed to preserve the
privacy in the known ciphertext model. The proposed system
MSRQE consists of the following phases:
Setup
Our system initializes in this phase. The data owner
generates Secret Key. DO generate n-bit random vector S
and two n×n invertible matrices M1, M2, Hence the secret
key SK is a form of three tuples SK = {M1, M2, S}.
GenerateIndex (DC, SK)
Data owner extracts keywords from document collection and
creates a dictionary with keywords and synonyms of each
keyword respectively, and then creates an index vector for
each document. DO construct unencrypted index tree by
742
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 1 (2016)pp 738-743
© Research India Publications. http://www.ripublication.com
Next, d10, d9, and d6 are reached with scores less than the kth
score of RankedList. Finally, the search algorithm tries to
search in l11 but the score of l11 is less than kth score of
RankedList so stops the search process. The search algorithm
returns RankedList with top four ranked documents d15, d7,
d8, and d12. In this illustration, we identified that the
DFSTsearching algorithm takes sub-linear time complexity.
[9]
[10]
CONCLUSION AND FUTURE WORK
In this paper, we intend to provide feasible solutions for
multi-keyword synonym ranked query problems over
encrypted cloud data while preserving strict system wise
privacy in cloud computing paradigm. Our main
contributions in this paper are four aspects: the first one
multi-keyword search, the second synonym based search,
third similarity ranked search and the last is efficient data
retrieval with BMS tree and DFST searching algorithm. Our
example illustration further shows efficient and accurate top
k documents retrieval of proposed scheme with sub-linear
time complexity. The query results more efficient and
accurate when data user enters synonyms of predefined
keywords or keywords due to lack of exact knowledge about
the data. In the future, first we will implement this scheme in
commercial clouds and design the algorithm for dynamic
update of BMS tree such as insertion of new documents and
deletion of documents from BMS tree.
[11]
[12]
[13]
[14]
[15]
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
C. Wang, "Secure ranked keyword search over
encrypted cloud data",Distributed Computing
Systems,
2010
IEEE
30th
International
Conference,IEEE, (2010).
N. Cao,C. Wang, M. Li, K. Ren, and W.
Lou,"Privacy-preserving multi-keyword ranked
search over encrypted cloud data", INFOCOM,
2011 Proceedings IEEE, (2011).
C. Wang, Kui Ren, Shucheng Yu, Urs, K.M.R
"Achieving usable and privacy-assured similarity
search over outsourced cloud data", INFOCOM,
2012 Proceedings IEEE, (2012).
C.Yang,Weiming Zhang, Jun Xu, JiajiaXu"A Fast
Privacy-Preserving Multi-keyword Search Scheme
on Cloud Data",Cloud and Service Computing
(CSC), 2012 International Conference,IEEE,
(2012).
W.K. Wong, D.W. Cheung, B. Kao, and N.
Mamoulis, “SecurekNN Computation on Encrypted
Databases,” Proc. 35th ACM SIGMOD Int’l Conf.
Management of Data (SIGMOD), pp. 139-152,
2009.
M.Chuah and W. Hu,"Privacy-aware bedtree based
solution for fuzzy multi-keyword search over
encrypted data",Distributed Computing Systems
Workshops,
2011
31st
International
Conference,IEEE, (2011).
S. Deshpande, "Fuzzy keyword search over
encrypted data in cloud computing",World Journal
of Science and Technology, vol. 2, no. 10, (2013).
D.X.Song,D. Wagner and A.Perrig,"Practical
techniques for searches on encrypted data, In
[16]
[17]
[18]
[19]
[20]
743
Security and Privacy", 2000. S&P 2000,IEEE,
(2000).
D. Zissis, D. Lekkas, Addressing cloud computing
security issues, Future Gener.Comput. Syst. 28 (3)
(2012) 583–592.
C. Wang, N. Cao, K. Ren, and W. Lou, “Enabling
secure and efficientranked keyword search over
outsourced cloud data,” IEEE Transactions on
Parallel and Distributed Systems, vol. 23, no. 8, pp.
1467–1479, 2012.
B. Wang, S. Yu, W. Lou, and Y. T. Hou, “Privacypreserving multi-keyword fuzzy search over
encrypted data in the cloud” in Proceedings of
INFOCOM. IEEE,2014.
N. Cao, C. Wang, M. Li, K. Ren, and W. Lou,
“Privacy-preserving multi-keyword ranked search
over encrypted cloud data,” IEEE Transactionson
Parallel and Distributed Systems, vol. 25, no. 1, pp.
222–233, 2014.
Christopher D. Manning, PrabhakarRaghavan, and
H. Schütze, Introduction to information retrieval.
Cambridge
university
press,
Cambridge,England2009, vol. 1.
P. D. Morehead, The New American Roget’s
College Thesaurus in Dictionary Form, 3rd ed.,
Signet Press: Seattle, 2002, pp. 10-900.
S. Yu, C. Wang, K. Ren, and W. Lou, “Achieving
Secure, Scalable,and Fine-Grained Data Access
Control in Cloud Computing,” Proc. IEEE
INFOCOM, 2010.
S. Kamara and K. Lauter, “Cryptographic Cloud
Storage,” Proc. 14th Int’l Conf. Financial
Cryptography and Data Security, Jan. 2010.
Zhihua Xia, Xinhui Wang, Xingming Sun, and Qian
Wang, “A Secure and Dynamic Multi-keyword
RankedSearch Scheme over Encrypted Cloud
Data,” IEEE Transactions on Parallel and
Distributed Systems, 2015.
A.Singhal, “Modern Information Retrieval: A Brief
Overview,” IEEE Data Eng. Bull., vol. 24, no. 4, pp.
35-43, Mar. 2001.
I.H. Witten, A. Moffat, and T.C. Bell, Managing
Gigabytes: Compressing and Indexing Documents
and Images. Morgan Kaufmann Publishing May
1999.
Veerraju
Gampala,
SrilakshmiInuganti,
SatishMuppidi, “ Data Security in Cloud Computing
with Elliptic Curve Cryptography,”International
Journal of Soft Computing and Engineering
(IJSCE),Volume-2, Issue-3, July 2012.