Multiple user decision making is important in to-day’s location-based service scenarios.Existing query services such as kNN and Skyline queries only consider single user and do not consider user’s preferencesThis system is designed using an authenticated query processing framework based on MR-tree.
Comprehensive experiments and Researchers have tested the effectiveness and robustness of this query and its 100% functional
2. CONTENTS
KEY WORDS.
WHAT IS IT?.
ADVANTAGES OF PROPOSED
AMUSQ.
HOW IT WORKS.
THE PROBLEM WITH EXISTING
QUERIES.
ALREADY EXISTING QUERIES.
OTHER AUTHENTICATED
QUERY PROCESSING.
AUTHENTICATION OF SPECIAL
CASES.
NOTATIONS.
CRITERIA FOR DORMINACE.
SIGNING.
MR TREE.
CONDITIONING.
QUEUING.
OVERVIEW OF TRANSVERSAL.
CONDITIONS FOR ACCEPTING RESULT.
THE CONTAINED ITEMS FOR EVERY (VO).
ALGORITHM 1(QUERY PROCESSING).
ALGORITHM II(AUTHENTICATING).
CONCLUSION.
4. WHAT IS IT?
Multiple user decision making is important in to-day’s location-based
service scenarios.
Existing query services such as kNN and Skyline queries only consider
single user and do not consider user’s preferences
This system is designed using an authenticated query processing
framework based on MR-tree.
Comprehensive experiments and Researchers have tested the
effectiveness and robustness of this query and its 100% functional.
5. CTd. WHAT IS IT?
It allows multiple users to define their own preferences because
it has more flexibility.
The query results are affected by spatial/non-spatial attributes
as well as users’ specified preferences.
6. THE PROBLEM WITH THE EXISTING
QUERIES
Existing query services
such as kNN and Skyline queries only consider single
user and their algorithms
it aims to process the query efficiently but not
considering the authentication problem.
8. ADVANTAGES OF PROPOSED AMUSQ
SOUNDNESS -every data item in the results comes from the original
database and is not tampered with by Service Provider.
CORRECTIVENESS -every data item in the results satisfies users’ query
criteria.
COMPLETENESS -every data item in the original database that satisfies
the query criteria is included in the results.
Lastly this method is applied into multiple users decision making
scenarios, under which users form a group and query for the best
answers according to their locations and preferences.
9. HOW IT WORKS
The figure in the next slide will show us how the authenticated query processing
framework under data outsourcing scenarios works.
A data owner builds and signs an Authenticated Data Structure (ADS) for his/her
dataset when he/she delegates it to a Service Provider (SP). The Service Provider
(SP) processes a user’s query by returning the corresponding results and a
Verification Object (VO). On receiving these information, the user recon-structs
querying results proof, and verifies their soundness, correctness, and
completeness with the published data owner’s signature
11. HOW IT WORKS
As a typical example of MUSQ, Alice and Bob query for the best dinner
place considering distance, customers’ ratings and prices.
Each has his/her own preference about these attributes.
For example, Alice prefers distance most but does not care about prices
or customer ratings, while Bob prefers prices and customers’ ratings
more than distance.
12. HOW IT WORKS
SP needs to retrieve the best dinner place satisfying both of their
requirements. A real application relevant to MUSQ is Baidu JuJu
introduced by Baidu company. As a highlight feature, Baidu JuJu makes
recommendation about the meeting spot for the user group based on
their locations.
However, it takes no consideration about the non-spatial attributes and
users’ needs. If MUSQ is used in this scenario, different users’
requirements can be satisfied and the meeting spot can be selected
more.
13. OTHER AUTHENTICATED QUERY
PROCESSING
Digital Signature Chaining – It was proposed to guarantee the
completeness of a dataset. It signs on adjacent data value, so that the
signature of each value depends both on its own value and its
immediate neighbor values.
Merkle Hash Tree (MHT) -is an effective Authenticated Data Structure
(ADS) to authenticate a large set of data values. MHT is a binary tree that
stores data in its leaf node.
Each leaf node has a digest generated by a one-way hash function H(·).
Each internal node has a digest of the concatenation of its children’s
digest. Only the root digest is signed by the data owner.
14. AUTHENTICATION OF SPECIAL CASES
Authentication for Skyline Queries- A skyline query finds the optimal POIs
for the user con-sidering both spatial and non-spatial attributes.
We can treat a skyline query as a special case of MUSQ when U = 1, W =
, and k is not defined.
SP follows the same query process as MUSQ, except when calculating
vsum, no weights are needed.
So the concept of weighted dominance no longer exists. SP only needs to
determine the dominance relation-ships among the visited MBRs or POI
objects during query processing.
15. CTd. AUTHENTICATION OF SPECIAL CASES
Authentication for kNN and kANN Queries- When users only provide their
preferences on the spatial attribute the MUSQ can be considered as k-
aggregated nearest neighbor (kANN) queries.
As a result, when dealing with dominance relationship between two objects,
only the aggregate distances are compared.
The single parameter comparison results in a single POI results that dominates
all the remaining POIs. So k rounds are needed to find k aggregated nearest
neighbors. The VO construction follows the same process of the original queries.
A kNN query returns k POI objects that are nearest to the querying user. It can
be considered as a special case of kANN when user set U contains only one
member.
16. CTd. AUTHENTICATION OF SPECIAL CASES
Query processing remains the same, except the spatial attribute is
directly represented by the distance between the querying user and the
POI object.
To authenticate the result, user also follows the process as:
1. Compare the root digest
2. Examine the dominance/weighted-dominance relationship of Gi and V
Oi .
3. Verify that POIs in the higher ranked groups dominate/weighted-
dominate those in the lower ranked ones.
17. NOTATIONS
Symbol
Meanin
g
ui a user’s identifier
P the dataset of POIs
p a POI in P
W the weight set of U
w the weight of a user
♦(p)
the weighted sum of non-spatial attributes of
p
adsum(p) the weighted sum of distance from p to U
p, u Euclidean distance from p to u
R the query result set
•a group of querying users
TABLE III
VALUE STORAGE TABLE
Symbol
Meanin
g
ui a user’s identifier
P the dataset of POIs
p a POI in P
W the weight set of U
w the weight of a user
♦(p)
the weighted sum of non-spatial attributes of
p
adsum(p) the weighted sum of distance from p to U
p, u Euclidean distance from p to u
R the query result set
•a group of querying users
TABLE III
VALUE STORAGE TABLE
Symbol
Meanin
g
ui a user’s identifier
P the dataset of POIs
p a POI in P
W the weight set of U
w the weight of a user
♦(p)
the weighted sum of non-spatial attributes of
p
adsum(p) the weighted sum of distance from p to U
p, u Euclidean distance from p to u
R the query result set
•a group of querying users
TABLE III
VALUE STORAGE TABLE
Symbol
Meanin
g
ui a user’s identifier
P the dataset of POIs
p a POI in P
W the weight set of U
w the weight of a user
♦(p)
the weighted sum of non-spatial attributes of
p
adsum(p) the weighted sum of distance from p to U
p, u Euclidean distance from p to u
R the query result set
•a group of querying users
TABLE III
VALUE STORAGE TABLE
Symbol
Meanin
g
ui
a user’s identifier
P the dataset of POIs
p a POI in P
W the weight set of U
w the weight of a user
♦(p) the weighted sum of non-spatial attributes of p
adsum(p) the weighted sum of distance from p to U
p, u Euclidean distance from p to u
R the query result set
•a group of querying users
18. CRITERIA FOR DORMINACE
if p and p_ satisfy the condition that (1) they cannot dominate each
other, (2) the ♦(p_) is no larger than ♦(p), then we say p_
weighted-dominates p on all non-spatial attributes.
19. SIGNING
h = hash(MBR1|h1|MBR2|h2| ・ ・ ・ |MBRn|hn)
Only the root node is signed by DO as Sig(hroot) using its private key.
The leaf entry e3 stores the MBR N1, the pointer to POIs p1, p4 in N1, and the digest derived from p1, p4 as
h(e3) = hash(p1|p4).
The internal entry e1 stores the MBR N3, the pointer to the child MBRs N1,N2, and the digest
derived from N1,N2 as h(e1) = hash(N1|h(e3)|N2|h(e4)).
The root node digest is h(eroot) = hash(N3|h(e1)|N6|h(e2)).
The root signature is generated by signing the digest of the root node eroot using the DO’s private key
21. CONDITIONING
• SP computes an additional value
vsum for each node N of the MR-tree and each POI object p in the leaf node.
• vsum will be used to determine the traversal sequence of the MR-tree
vsum(p) = adsum(p) + ♦(p)
• vsum(N) = min pi∈N (adsum(N.pi)) + min pj∈N (♦(N.pj))
• we must have vsum(p) < vsum(p_); if a POI p dominates or weighted dominates a
region S
22. QUEUING
The root node of MR-tree is added to H and VO at first
For each step, the top of H pops up,
SP examines whether the popped entry of H is dominated
or weighted-dominated by any existing objects in R.
the corresponding node or object is added into VO
23. OVERVIEW OF TRANSVERSAL
if it’s an internal node of MR-tree, its child nodes will
be added into H
if it’s a leaf node, all the POI objects covered by the
node will be added into H
if it’s a POI object, it will be added directly into R
24.
25. CONDITIONS FOR ACCEPTING RESULT
If _R_ > k, the number of query result exceeds users’ requirements, They
mark the top-k POIs as the final result according to their values of
vsum, then adds the rest into VO
If _R_ = k, it means the result set can exactly meet users’ requirements
If _R_ < k, a second round query is needed because the number of
result set from the first round query cannot satisfy users’ preferences
k is an optional parameter. If it’s not defined by U, SP should return all
the calculated result objects of the first round in R.
26. THE CONTAINED ITEMS FOR EVERY (VO)
(1) The POI objects,
(2) The MBRs along with theirs digests, and
(3) All users’ query information U and W.
29. CONCLUSION
This scheme enables the query processing to generate
results and proofs that to check each result is sound, correct
and complete.
Through theoretical proof and performance evaluation, it
shows that this framework is not only feasible, but also
efficient and robust under various parameter settings.