A Framework for Expressing and Combining Preferences

1
A Framework for Expressing and Combining
Preferences

• Rakesh Agrawal, Edward L. Wimmers
• Presented by Zhen Zhang

2
A Big Gap
Select
I Prefer
150 lt Price lt 170 and near campus without
parking 150 lt Price lt 200 and near campus
without parking 150 lt Price lt 250 and near bus
line without parking
Price 150 250350 Location near campus
doesnt matter Size doesnt matter 80
above Parking No Included All kinds of
features ..
3
So Many Troubles

• Issue so many queries
• Relax and narrow down
• Browse so many results
• Cannot sort by my preference
• So many apts near campus, sort by the closeness
  of price to 150

4
Framework

• Provide a way for user to express the preferences
• Provide a mechanism to tradeoff
• Different factors
• Different users

5
Relational Data Model

• Each object has set of attributes
• apt ( price, location, size, parking, )
• Preference is defined on the attributes, and
  quantified into scores
• Price within 150 10 1.0
• Different user may define preferences on
  different subset of attributes
• Me price, location, parking
• Roommate price, size, parking
• The combination function is used to negotiate
  between different preferences

6
Express the Preference

• Somehow like conjunction query

7
Express the Principles of Tradeoff

• Resolve the conflict
• Between two preferences
• Within the preference
• (150, 80, On campus, No Parking)

8
Combination Function

• Implement the principle of how to trade off and
  negotiate
• Combine different preferences and reach to a
  final decision

9
Combination function
10
Combination Function

• Input
• All the possible scores an object can get
• (0.9, 1.0, No)
• The content of object
• (150, 80, On campus, No Parking)
• Output
• The preference score of object

11
An Example
Combine(myscore,herscore, car) if price lt 200
max from herscore else average from
myscore
(150, 80, On campus, No Parking) Myscore 0.9,
1.0 Herscore No) Final score No
12
Nice Properties

• Autonomy
• Work individually and independently
• Modular
• Localize the change
• Its a Closure!
• Complicated preference can be built incrementally

13
Conclusion

• Framework for expressing and combining
  preferences
• Preferences are combined in a flexible way
• Each preference keep its autonomy

14
Discussion Question

• Does this framework make our life easier?
• Easy to use
• Expressiveness, or powfulness

15
Discussion Quesion

• Are there other way to express preferences? In
  terms of scores? Language?

16
Discussion Question

• Whats different from previous paper in
  expressing and combining preference? Which is
  more feasible?
• Can we adopt the algorithm from previous work
  here? Or you may propose better implementation?

17
Discussion Question

• Is modular good (as the author claimed) or bad,
  in what sense?