Davide Mottin (U of Trento), A. Marascu (IBM Research - Ireland), S. Basu Roy (U of Washigton Tacoma), G.
Das (U of Texas Arligton), T. Palpanas (Paris Descartes U) and Y. Velegrakis (U of Trento)
Full technical details in: Davide Mottin, A. Marascu, S. Basu Roy, G. Das, T. Palpanas and Y. Velegrakis, "A Probabilistic Optimization
Framework for the Empty-Answer Problem", Proceedings of VLDB, 6(14), 2013
0
0
0
0
0
1
0
1
0
1
0
1
The Theory
Turbo
1
1
0
0
ESP
DSL
4WD
0
0
0
0
HiFi
0
1
0
0
Alarm
MP3
ABS
VW Touareg $62K 1
Askari
A10 $206K 0
Honda Civic $32K 1
Porsche
911
$126K 0
Manual
t1
t2
t3
t4
Price
Make
Model
A Database
 For a relaxation Q’ of Q
1
1
0
1
0
0
0
0
 Prior
Belief of the user that an answer will be found in the database
 Prefer
The likelihood the user will like the relaxed query answers
A Query
 Relaxation Preference Function
 Probability to reject the a relaxation
Cars with ABS, DSL & Manual Transmission
The Answer: None
 Probability to accept the relaxation
What Users Want
Interactive
100%
80%
60%
40%
20%
0%
Favored
Multi-Relaxations
Answers Quality
top-k
Why-Not
 Cost for a relaxation
Usability
The Relaxation Tree
 Precompute the whole Relaxation Tree
 Fast Optimal
 Start from the root and construct the tree on
demand, computing min and max cost bounds
 Approximate (CDR)
 Approximate the cost of each relaxation
10
10000
1000
100
10
1
0.1
0.01
0.001
FullTree
Query time (sec)
 Full Tree
The Performance
Query time (sec)
The Algorithms
CDR
FastOpt
3
4
5 6 7 8
Query size
9 10
FullTree
FastOpt
CDR
8
6
4
2
0
0
100 200 300 400 500
Number of tuples (k)