Working with Preferences

1
Working with Preferences

• Ronen Brafman
• Computer Science Department
• Ben-Gurion University

2
Goal of our Work Provide tools and build systems
that can make, or can help users make, good
choices. Such systems must understand the
users preferences to be useful.
3
Applications of Interest

• optimal product/item selection
• Select a flight, a camera, a movie,
• optimal product configuration
• Configure a PC, a vacation, a car,
• personalization
• Personalize content, interface,
• guiding program choices
• Program will make choices based on a preference
  model provided by the designer
• We target lay users and application designers
  with no background in decision theory

4
Focus of this Talk Provide an overview of an
approach to preference handling based on the
xCP-net models and algorithms, and discuss some
applications
5
Constrained Optimization Process
Manufacturer Constraints
Feedback ? Spec Revision
Updated Choice
6
The Key Ideas

• Preference elicitation
• Use natural language like statements make it
  simple
• Support heterogeneous types of statements
• Make scaling up to quantitative assessment
  possible
• Calculating optimal choices
• Efficient methods for ordering sets
• Algorithms for constrained optimization
• Efficient data-driven incremental elicitation

7
Elicitation and Modeling
8
Natural Preferences Statements

• Value preferences
• Strict I prefer an isle seat to a window seat
• Conditional I prefer isle to window in economy
  class
• Attribute importance
• Strict Seat assignment is more important than
  airline
• Conditional Seat assignment is less important
  than airline in domestic flights

9
The Ceteris Paribus Semantics

• Ceteris Paribus Latin all else being equal
• I prefer an isle seat to a window seat ?
• Given two flights that differ only in seat
  assignment, I prefer the one with an isle seat.
• Otherwise, cant infer a preference
• Seat assignment is less important than airline
  in domestic flights ?
• Given two domestic flights that differ in seat
  assignment and airline only, I prefer the one
  with a better airline
• Says nothing about two international flights

10
CP-nets (Boutilier, Brafman, Hoos, Poole 99,
Boutilier, Brafman, Domshlak, Hoos, Poole 04)

• A qualitative, graphical model of preferences,
  that captures and organizes statements of
  conditional value preference.
• Each node represents a domain variable.
• Parents(v) are those variables that affect users
  preference over the values of v
• Parents(class) airline
• Conditional preference table (CPT) associated
  with every node in the CP-net
• Provides an ordering over the values of the node
  for every possible parent context

11
Example of a CP-net
12
Semantics and Consistency
Any acyclic CP-net defines a (consistent)
partial order over the outcome space.
13
Importance Relations
14
More Complex Example
Day of the flight
Airline
Departure Time
Stop-overs
Class
15
Day of the flight
Departure Time
Airline
Stop-over
Class
16
Day of the flight
Departure Time
Airline
Stop-overs
Class
17
nodes ? variables
cp-arcs (directed)
A
i-arcs (directed)
ci-arcs (undirected)
B
E
cp-tables
ci-tables
C
D
18
Role of Graphical Structure

• Users need not be aware of the underlying
  graphical structure
• User employs statement templates or some other
  input interface
• System constructs the network on the fly
• Graphical structure important for
• Analysis query complexity related to structure
• Algorithms often use topological sort

19
Preferences ? Plausibility

• CP Conditional Plausibility (Jerome Lang)
• p is more plausible than p given q (ceteris
  paribus)
• Given two worlds satisfying q that are identical
  except for the value of p, the one satisfying p
  is more plausible than the one satisfying p.
• p is more important, plausibility wise, than q
  (ceteris paribus)
• Worlds in which p has its more plausible value
  are more plausible than worlds in which q has its
  more plausible value

20
UCP-Nets Boutilier, Bacchus, Brafman

• Quantified CPT
• Simplified semantics sum of utility factors
  U(abcd)
• fA(a)fB(b)fC(abc)fD(cd)
• 5 4 .6 .9 10.5
• Linear time computation
• Linear time comparison

a a 5 2
b b 4 3
c c
C
a b .6 .1 a b .2 .8 a b .3 .8 a b .9 .3
d d
c .9 .8 c .2 .3
D
21
Compiling Diverse Information into UCP-nets

• Diverse statement can be compiled into UCP-net
• Independence information maintained by graph
• Value preferences and variables preferences
• Two-way comparisons of complete outcomes
• Quantitative information
• Simple compilation based on linear constraints
• Good for incremental refinement
• Start with qualitative model no cold start!
• Refine with user feedback/additional observations

22
Using (T)CP-nets Algorithms

1. Selecting the best element
2. Ordering and constrained optimization

23
1. Ordering Elements

• Need Order (the many) results of a query
• Naïve solution pairwise comparison -- ogto ?
• Problem NP-hard in general
• Alternative solution linearize the partial order
• Insight some linearizations are easy to obtain
• Key
• One of oeo or oeo can be decided quickly
• Answers used to generate consistent ordering

24
1. Preferential Optimization
Finding the preferentially optimal outcome for an
acyclic network is straightforward!
25
2. Preference-Based Constrained Optimization
(PCO)

• Given
• User preferences
• An implicitly specified set of feasible options
• Find one/all/k optimal, feasible element(s)
• Applications
• Product configuration (PC, vacation)
• Optimal plan selection
• Content/display adaptation and personalization

26
Solving PCOOrdered Generate Test

• Generate outcomes in non-increasing order
• linearize the partial order over all possible
  elements
• Test for feasibility
• Check for optimality
• First feasible outcome is optimal!
• Need more?
• Maintain set of optimal solutions
• New solution is optimal if not dominated by any
  previously generated solution

27
Generating a Non-increasing Sequence of Outcomes

• Topologically sort the variables
• Build an assignment (search) tree by
  instantiating variables according to this order
• Order variable values based on the CPT
• Leaf nodes, ordered left to right ( depth-first
  traversal of the tree) correspond to a
  non-increasing sequence of outcomes

28
(No Transcript)
29
Sorting?

• We just saw how we can order all outcomes
• This method does not work for sorting a subset of
  all possible assignments
• Fortunately, there is a method that can be used
  to sort n outcomes in time O(n log n)

30
Ordered Generate and Test ? Efficient Constrained
Optimization

• Tree search favorite CSP pruning techniques
• Equivalent to solving the CSP with meta-level
  constrains on variable and value ordering
• Branch and bound eliminate sub-tree when
• We assign a variable to a less preferred value
• Current set of constraints as strong as for some
  previous value of this variable

31
(No Transcript)
32
Anytime Behavior

• First feasible solution is optimal!
• No theoretical overhead beyond standard CSP
  solution
• No item withdrawn from set of current solutions
• To obtain more than one solution dominance
  testing required
• Can lead to considerable computational overhead

33
Applications
34
Flight Selection (Brafman, Domshlak, Kogan
UAI04)

• Instance of selecting optimal element problem
• Approach
• User supplies
• constraints (source, destination, etc.)
• initial preferences
• Preferences compiled into UCP-net
• Top k results shown
• User provides feedback
• Which flight is most preferred among k best
• Any new preferences
• Top results revised accordingly

35
Planning for Data Products (Golden et. al.)

• NASA collects much raw data about earth
• Requires extensive processing to be useful
• Each Earth scientists needs different processing
• ImageBot generates data plans for such products
• Data plans run scientific models, combine and
  transform data in order to achieve data goals
• Many plans can generate one data product
• Each plan has different value
• Using a simple preference language ImageBot
  planning algorithm is biased towards more
  preferred plans

36
Adaptive, Personalized Rich Media Presentations

• Presentations with diverse media elements
  requiring spatio-temporal synchronization
• Target wide audiences
• Audience members have different tastes, different
  network connections, and use different devices
• Goal provide designers with tools for designing
  presentations that adapt to user and user context
• Demo

37
Example ESPN Promo

• Presentation with five elements
• Video featuring upcoming broadcast
• 2 image ads
• Video ad
• Running text with scores/news
• Each media element is a variable
• Additional variables denote user properties,
  device properties, bandwidth
• Variable values different content and quality
  options

38
How Does It Work?

• Variables denote different presentation elements
  (video, ads, running text)
• Additional variables denote context information
• TCP-nets specify preference relation over
  presentations
• Author may specify additional constraints
• At download time combine
• TCP-net
• Information about user and user context
• Resulting constrained optimization problem solved
• Output a SMIL presentation

39
CP-Net for ESPN Promo
Gender
Nationality
40
Real-Time Content Selection for Command and
Control

• Imagine a decision maker monitoring masses of
  data in a real-time command control center for
  all rescue forces in Paris
• Video streams
• Sensor data
• Results of relevant queries
• Results of data analysis (e.g., simulations, risk
  assessment)
• Task Which data to show at each point in time?

41
Possible Information Sources

• Cameras on fireman helmet
• Fixed surveillance cameras
• Heat sensors, smoke detectors, co2-levels
• Area maps, building plans, driving distance,
  number of residents
• Simulation of structure strength, time to contain
  a fire as function of wind and other weather
  conditions, etc.
• Demo

42
Our Proposal Decision Theoretic Control

• Offline build a preference model capturing the
  value of different choices
• Online compute best choices
• In our case which data to show
• In general whatever choices the system must make
• Not new, but not too practical, so far
• Main obstacle preference handling

43
What Are the Challenges

• ESPN solution (using CP-nets) is static
• We know all the variables and their possible
  values at specification time
• Our context is dynamic
• Different cities will have different relevant
  information streams
• Same city will have different relevant
  information streams at each time
• A single specification should handle LA, Paris,
  NY, etc.
• Our solution relational preference rules
• A relational model that generalizes of UCP-nets
• Assumes concepts/classes are fixed (fireman,
  fire, fire truck)
• Does not assume anything about specific instances

44
Modeling a Fire Department

• Rules
• Fireman(x) ? fire(y) ? x.location y.location
• ? x.camera.display on 4, off 0
• Fireman(x) ? x.co2-levelhigh
• ? x.camera.display on 8, off 0

45
Other Interesting Issues

• Set preferences specifying and computing
  preferred subsets when items have synergy
  AAAI06,AAAI07
• Browsing large heterogeneous databases
• Joint work with Scott Klemer, Ron Yeh, and Yoav
  Shoham supported by NSF

46
Collaborators

• Carmel Domshlak Technion
• Craig Boutilier University of Toronto
• Holger Hoos, David Poole UBC
• Doron Friedman University College London
• Solomon Shimony Ben-Gurion University

47
Summary

• Much recent interest in work on preference
  representation, elicitation, and reasoning
• xCP-nets offer a simple preference language and
  convenient graphical and algorithmic tools
• UCP-nets provide good target for knowledge
  compilation
• Many applications in e-commerce and user
  interfaces

48
(No Transcript)