Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

30
Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs Tyler Lu and Craig Boutilier University of Toronto

description

Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs. Tyler Lu and Craig Boutilier University of Toronto. Introduction. New communication platforms can transform the way people make group decisions. - PowerPoint PPT Presentation

Transcript of Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Page 1: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost

Tradeoffs

Tyler Lu and Craig BoutilierUniversity of Toronto

Page 2: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

IntroductionNew communication platforms can transform

the way people make group decisions.

How can computational social choice realize this shift?

ChoicesPeople

Computational Social Choice

Consensus

2

Page 3: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Introduction• Computational social choice

– Aggregate full preferences (rankings)– Mostly study rank-based schemes (Borda, maximin, etc…)

• Rank-based voting schemes rarely used in practiceProblem: Cognitive and communication burdenOur approach (recent work): Elicit just the right preferences to make good enough group decisionsThis work: Multi-round elicitation and probabilistic preference models to further reduce burdensAlice Bob Cindy

>12

1

3

Page 4: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Outline

• Preliminaries

• Multi-round Probabilistic Vote Elicitation

• Methodology and Analysis for One-round

• Experimental Results

4

Page 5: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Preliminaries• Voters N = {1..n}; alternatives/items A = {a1…am}

• Vote vi is a ranking of A

• Complete profile v = (v1, …, vn)

Alice Bob1

23

voting rule r

5

Cindy

Page 6: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Score-based Rules

• Many rules have score-based interpretation– Surrogate for “total group satisfaction”– E.g. Borda, Bucklin, maximin, Copeland, etc…

• Associates a score for each item given full rankings s(a, v)

• Winner has highest score6

s( , v) = 7

s( , v) = 6

s( , v) = 5

Alice Bob1

23

Cindy Borda scores

Page 7: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Partial Preferences

• Partial vote pi is a partial order of A– Represented as a (consistent) set of pairwise comparisons– Higher order: top-k, bottom-k, …– Easy for humans to specify

• Partial profile p

7

Alice

>

> >

How to make decision with partial preferences?

Page 8: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Decision with Partial Preferences

• Possible and necessary co-winners [Konczak, Lang’05]

• Recently: minimax regret (MMR) [Lu, Boutilier’11]

– Provides worst-case guarantee on score loss w.r.t. true winner

– Small MMR means good enough decision– Zero MMR means decision is optimal

8

Page 9: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Minimax Regret

9

Adversarial

Bestresponse

Page 10: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Vote Elicitation

• MMR: good choices with “right” partial votes– How to minimize amount of partial preference

queries to make good decision?

• MMR-based incremental elicitation [Lu, Boutilier’11]

– Problem: must wait for response before next query

10

Page 11: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Incremental Elicitation Woes

• Each query is a (voter, pairwise comparison) pair– Exploits MMR, depends on all previous responses

11

ElicitorYES NO

…> ? > ?…

Bob annoyed at having to come back to answer query“interruption cost”

Page 12: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Our Solution:Multi-Round Batching

• Send queries to many voters in each round

12

Elicitor Round: 1Give your top 2

1. 1. 1.

Round: 2Give your next top 1

3. 3. 3.

MMR ≤ εRecommendation:

Interruption cost reduced

2. 2. 2.

Page 13: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Multi-Round Probabilistic Vote Elicitation

• Query class: “rank top-5”, “is A > B?”, etc…– Single request of preferences from voter– Have different cognitive costs

• In each round π selects a subset of voters, and corresponding queries– Can be conditioned on previous round responses

• Function ω, selects winner and stops elicitation• How to design elicitation protocol with provably good

performance?– Worst-case not useful (for common rules)– Use probabilistic preference models to guide design

13

Page 14: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Multi-Round Probabilistic Vote Elicitation

• Distribution P over vote profiles– Induced distribution over runs of protocol (π, ω)

• Can define distribution over performance metrics

14

Quality of winner: Max regret, expected regret

Amount of information elicited: equivalent #pairwise comparisons, or bits.

Number of rounds of elicitation

Tradeoffs!

Depends on what costs are

important.

Page 15: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

One-Round Protocol

• Query type: top-k– “Rank your top-k most preferred”

• Simple top-k heuristics [Kalech et al’11]

– Necessary and possible co-winners– No theoretical guarantees on winner quality– Don’t provide guidance on good k– No tradeoff between winner quality and k

15

Page 16: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Probably Approximately Correct (PAC) One-Round Protocol

• Any rank-based voting rule• Any distribution P over profiles

• What is a good k?– p[k] are partial votes after eliciting top-k

k*: smallest k, with prob. ≥ 1 - δ, MMR(p[k]) ≤ ε

• As long as we can sample from P, we can find “approximately” good k…– Samples can come from historical datasets, surveying, or generated

from learned distribution

16

Page 17: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Probably Approximately Correct One-Round Protocol

General Methodology• Input: sample of vote profiles: v1, …, vt

• MMR accuracy ε > 0• MMR confidence δ > 0• Sampling accuracy ξ > 0• Sampling confidence η > 0

Find best the smallest k with

17

Page 18: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Probably Approximately Correct One-Round Protocol

Theorem: if sample size

then for any P, with probability 1 - η, we have

(a) ≤ k* (b) P[ MMR(p[ ]) ≤ ε ] ≥ 1 - δ - 2ξ

18

Page 19: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Practical Considerations

• Sample size from theorem typically unnecessarily large

• Empirical methodology can be used heuristically

• Can generate histograms of MMR for profile samples from runs of elicitation– Can “eyeball” a good k– Can “eyeball” tradeoffs with MMR

19

Page 20: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Experimental Results

• First experiments with Mallows distribution– Rankings generated i.i.d.– Unimodal, with dispersion parameter– t = 100 profiles (for guarantees, use bounds for t)

• Borda voting• Simulate runs of elicitation– Measure max regret and true regret– Normalize regret by number of voters

20

Page 21: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Experimental Results

21

x-axis is MMR per voter

Page 22: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Experimental Results

22

Page 23: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Experimental Results

23

Page 24: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Experimental Results

24

Page 25: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Experimental Results

25

Page 26: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Experimental Results

26

Sushi 10 alternatives50 profiles, each with 100 rankings

Page 27: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Experimental Results

27

Dublin North12 alternatives73 profiles, each with 50 rankings

Page 28: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Concluding Summary

• Model of multi-round elicitation protocol– Highlights tradeoffs between quality of winner,

amount of information elicited, and #rounds– Probabilistic preference profiles to guide design and

performance instead of worst-case• One-round, top-k elicitation– Simple, efficient empirical methodology for choosing k– PAC guarantees and sample complexity– With MMR solution concept, enables probabilistic and

anytime guarantees previous works cannot achieve

28

Page 29: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

Future Work

• Multi-round elicitation, top-k or pairwise comparisons

• Fully explore above tradeoffs (associative different costs)

• Assess expected regret and max regret

29

Page 30: Vote Elicitation with Probabilistic Preference Models: Empirical Estimation and Cost Tradeoffs

The End

30