1. Elicitation of Constraints and Qualitative
Preferences in Multi-Attribute Auctions
MSc Thesis Defense
Shubhashis Kumar Shil
26 November 2013
1
2. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
1. Introduction
2. Proposed MARA Protocol
3. Experiments and Evaluation
4. Conclusion and Future Work
1.1 Problem Statement
1.2 Motivations
1.3 Contributions
Outline
2
2.1 Constraint Elicitation
2.2 Preference Elicitation
2.3 Weight Calculation Automation
2.4 Utility Function Calculation Automation
2.5 Bid Evaluation
3. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
Eliciting the buyer’s requirements, which consists of constraints and
qualitative preferences, adequately
Determining the winner, which has been shown to be computationally
complex, efficiently according to the buyer’s requirements
1.1 Problem Statement
3
4. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
Often, the buyer is comfortable to express his preferences about the product
qualitatively
There should be options for the buyer to specify constraints
The constraints and preferences can both be non-conditional or conditional
It is more efficient for the system to process quantitative data
Provide the buyer with more comfort as well as keep the system efficient
1.2 Motivations
Prefer expressing “Brand attribute is very much important” to “Importance of
Brand attribute is 80%”
4
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
5. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
Develop MARA protocol:
Enhance MAUT:
1.3 Contributions
Allowing the buyer to express his qualitative non-conditional and conditional
preferences
Allowing the buyer to specify non-conditional and conditional constraints
Allowing the constraints and preferences co-exist in the system
Assisting both the buyer and sellers with friendly graphical user interfaces
Designing a 3-layer software architecture based on multi-agent technique and
Belief-Desire-Intention (BDI) model
Converting qualitative requirements into quantitative ones
Automating the MAUT calculation
Determining the winner efficiently
5
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
6. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions 10
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
Constraint Elicitation
A process to extract hard constraints from a user that must
be satisfied completely
where
(condition1) and/or, ..., and/or (conditionn) => constraint
conditioni : rel (attribute, value of attribute)
constraint : rel (attribute, value of attribute)
rel ε {=, ≠, <, >, ≤, ≥}
7. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions 10
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
Preference Elicitation
A process to extract preferences as soft constraints from a user
that are considered as wishes or desires
where
(condition1) and/or, ..., and/or (conditionn) => preference
conditioni : rel (attribute, value of attribute)
preference : attribute (value1 (likeliness), ..., valuem (likeliness) )
rel ε {=, ≠, <, >, ≤, ≥}
8. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions 10
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
Automating Weight Calculation
Importance Level Quantitative
Importance Level
Extremely Important 1
Very Much Important 0.75
Important 0.5
Not Much Important 0.25
weightRatepLevelquanrankweight aaa Im
M
a
aweight
1
1/
1 aa positionMrank
where
M = Number of attributes
positiona = position of attribute a in the attribute list ordered by the buyer
9. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions 10
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
Automating Utility Function (String) Calculation
eutilityRatnessquanLikelirankvU aa vvaa )(
1 aa vv positionNrank
where
N = Number of values of an attribute
positionva
= position of attribute value va in the list of the
values of that attribute ordered by the buyer
Attribute
Value Type
Likeliness Quantitative
Likeliness
String Highest
Above Average
Average
Below Average
Lowest
1
0.8
0.6
0.4
0.2
Numeric Highest
Lowest
1
0.2
10. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions 10
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
Automating Utility Function (Numeric) Calculation
0)(;1)( laahaa vUvU
)/()()( lahalaaaa vvvvvU
)( laa vU Second lowest utility value/number of attribute values
where
Ua(vha) = Utility value of attribute value for the highest likeliness
Ua(vla) = Utility value of attribute value for the lowest likeliness
va = a value of attribute, a
vha = a value of attribute, a of highest likeliness
vla = a value of attribute, a of lowest likeliness
11. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
MAUT*
Buyer’s Request of Purchase
Buyer’s Preference Elicitation Buyer’s Constraint Elicitation
Qualitative Preference of
Attributes (Importance Levels)
Qualitative Preference of Attribute
Values (Likeliness)
Constraint Checking
Conversion of Attribute
Preferences
Conversion of Attribute
Value Preferences
Quantitative Preference of
Attributes
Quantitative Preference of
Attribute Values
Calculation of Attribute
Weights
Calculation of Attribute
Utility Function
Weight of Attributes Utility Function Value of Attributes
MAUT Calculation
Overall MAUT Utilities
Valid Bids
10
Requirements Elicitation and Bid Evaluation in MARA System
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
A Summary
12. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
Presentation Layer
Graphical User Interface Agent
Buyer’s
Preference
Elicitation
Buyer’s
Constraint
Elicitation
Seller
Bidding
Bid Score
and Status
Display
Business Logic Layer
Winner Determination Agent
Admin
Agent Constraint
Checking
Bid Evaluation
with MAUT*
Data Access Layer
Auction
Database
Product
Database
Presentation Layer
Business Logic Layer
Data Access Layer
Stores two databases: Auction
and Product
Graphical User Interface Agent
interacts with user
Winner Determination Agent
determines the winner with the
help of Admin Agent
12
MARA 3-Layer Software Architecture
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
13. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
BDI Model
Admin Agent
Graphical User Interface (GUI) Agent
Winner Determination (WD) Agent
15 plans
15 plans
18 GUI windows
2 plans
Jadex
Agent simulation framework
13
Jadex Control Center GUI
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
14. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
A reverse auction of a television consists of 10 attributes : Brand, Customer
Rating, Display Technology, Model Year, Price, Refresh Rate, Resolution, Screen
Size, Warranty and Weight
1 buyer and 20 sellers
4 non-conditional constraints, 3 conditional constraints, 7 non-conditional
preferences and 3 conditional preferences
14
Case Study
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
15. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
Non-Conditional Constraints (NCC)
(ncc1) NULL → Model Year ≠ 2011
(ncc2) NULL → Warranty ≥ 2
(ncc3) NULL → Refresh Rate ≥ 120
(ncc4) NULL → Screen Size ≥ [30 - 39]
Conditional Constraints (CC)
(cc1) (Refresh Rate ≤ 240) → Price ≤ [900 - 999.99]
(cc2) (Brand = Panasonic) and (Resolu on = 720p HD) → Weight ≤ [5 - 5.9]
(cc3) (Brand = LG) or (Resolu on = 1080p HD) → Screen Size ≤ [40 - 49]
15
Figure 6: Constraint Elicitation
Assisting the Buyer to Specify the Constraints via GUIs
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
17. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
Disqualified - Does not satisfy constraint(s) completely
Challenged - Satisfies constraints completely but the overall utility value is not the highest
Winner - Satisfies constraints completely and the overall utility value is the highest
Bid Submission, Evaluation & Status
17
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
19. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
Attributes Sellers NCC CC NCP CP Execution
Time
6 (A1-A6) 20 3
(ncc1, ncc3,
ncc5)
2
(cc1, cc5)
4
(ncp1, ncp2,
ncp3, ncp5)
0 0.203
7 (A1-A7) 20 3
(ncc1, ncc3,
ncc5)
2
(cc1, cc5)
4
(ncp1, ncp2,
ncp3, ncp5)
1
(cp2)
0.328
8 (A1-A8) 20 4
(ncc1, ncc3,
ncc4, ncc5)
3
(cc1, cc3,
cc5)
5 3
(cp1, cp2,
cp5)
0.437
9 (A1-A9) 20 5 4
(cc1, cc3,
cc4, cc5)
5 4
(cp1, cp2,
cp4, cp5)
0.438
10 20 5 5 5 5 0.453
Execution time increases with the increment of the number of attributes
19
Execution Time of MAUT* by varying Number of Attributes
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
20. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
Execution time increases with the increment of the number of sellers
Attributes Sellers NCC CC NCP CP Execution
Time
10 4 (S1-S4) 5 5 5 5 0.219
10 8 (S1-S8) 5 5 5 5 0.250
10 12 (S1-S12) 5 5 5 5 0.313
10 16 (S1-S16) 5 5 5 5 0.359
10 20 5 5 5 5 0.453
20
Execution Time of MAUT* by varying Number of Sellers
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
21. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
Execution time decreases with the increment of the number of
non-conditional constraints
With the increment of the number of non-conditional constraints, it creates
more chance for the bids to be disqualified
Attributes Sellers NCC CC NCP CP Execution
Time
10 20 1
(ncc1)
5 5 5 0.532
10 20 2
(ncc1, ncc2)
5 5 5 0.516
10 20 3
(ncc1, ncc2,
ncc3)
5 5 5 0.500
10 20 4
(ncc1, ncc2,
ncc3, ncc4)
5 5 5 0.469
10 20 5 5 5 5 0.453
21
Execution Time of MAUT* by varying Number of Non-Conditional Constraints
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
22. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
Execution time decreases or remains the same with the increment of the
number of conditional constraints
With the increment of the number of conditional constraints, it creates
more chance for the bids to be disqualified
Attributes Sellers NCC CC NCP CP Execution
Time
10 20 5 1
(cc1)
5 5 0.578
10 20 5 2
(cc1, cc2)
5 5 0.546
10 20 5 3
(cc1, cc2, cc3)
5 5 0.546
10 20 5 4
(cc1, cc2, cc3,
cc4)
5 5 0.484
10 20 5 5
(cc1, cc2, cc3,
cc4, cc5)
5 5 0.453
22
Execution Time of MAUT* by varying Number of Conditional Constraints
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
23. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
Execution time increases with the increment of the number of
non-conditional preferences
Attributes Sellers NCC CC NCP CP Execution
Time
10 20 5 5 1
(ncp1)
5 0.401
10 20 5 5 2
(ncp1, ncp2)
5 0.408
10 20 5 5 3
(ncp1, ncp2,
ncp3)
5 0.417
10 20 5 5 4
(ncp1, ncp2,
ncp3, ncp4)
5 0.437
10 20 5 5 5 5 0.453
23
Execution Time of MAUT* by varying Number of Non-Conditional Preferences
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
24. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
Execution time increases with the increment of the number of conditional
preferences
Attributes Sellers NCC CC NCP CP Execution
Time
10 20 5 5 5 1
(cp1)
0.313
10 20 5 5 5 2
(cp1, cp2)
0.407
10 20 5 5 5 3
(cp1, cp2, cp3)
0.421
10 20 5 5 5 4
(cp1, cp2, cp3,
cp4)
0.438
10 20 5 5 5 5 0.453
24
Execution Time of MAUT* by varying Number of Conditional Preferences
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
25. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
Our MARA protocol is able to elicit non-conditional and conditional
constraints
The system is able to elicit qualitative non-conditional and conditional
preferences
Our improved MAUT can take qualitative requirements and convert them
into quantitative ones
The system provides automation of the MAUT algorithm
The system can determine the winner efficiently
Conclusion
25
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work
26. Elicitation of Constraints and Qualitative Preferences in Multi-Attribute Auctions
Besides MAUT there are other techniques [9] such as Analytic Hierarchy Process
(AHP), Weight determination based on Ordinal Ranking of Alternatives (WORA)
and Simple Multi-Attribute Rating Technique (SMART) that can be used
The system can be tested with real world datasets of auction systems
Our MARA system can be improved by allowing the buyer to specify his
requirements qualitatively on some attributes and quantitatively on other
attributes of the product he is interested in
Future Work
26
Introduction Proposed MARA Protocol Experiments and Evaluation Conclusion and Future Work