interactively optimizing information retrieval systems as a dueling bandits problem

38
Interactively Optimizing Information Retrieval Systems as a Dueling Bandits Problem ICML 2009 Yisong Yue Thorsten Joachims Cornell University

Upload: ford

Post on 25-Jan-2016

51 views

Category:

Documents


1 download

DESCRIPTION

Interactively Optimizing Information Retrieval Systems as a Dueling Bandits Problem. ICML 2009 Yisong Yue Thorsten Joachims Cornell University. Learning To Rank. Supervised Learning Problem Extension of classification/regression Relatively well understood - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Interactively Optimizing Information Retrieval Systems as

a Dueling Bandits Problem

ICML 2009

Yisong Yue Thorsten Joachims

Cornell University

Page 2: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Learning To Rank

• Supervised Learning Problem– Extension of classification/regression– Relatively well understood– High applicability in Information Retrieval

• Requires explicitly labeled data– Expensive to obtain– Expert judged labels == search user utility?– Doesn’t generalize to other search domains.

Page 3: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Our Contribution

• Learn from implicit feedback (users’ clicks)– Reduce labeling cost– More representative of end user information needs

• Learn using pairwise comparisons– Humans are more adept at making pairwise judgments– Via Interleaving [Radlinski et al., 2008]

• On-line framework (Dueling Bandits Problem)– We leverage users when exploring new retrieval functions– Exploration vs exploitation tradeoff (regret)

Page 4: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Team-Game Interleaving

1. Kernel Machines http://svm.first.gmd.de/

2. Support Vector Machinehttp://jbolivar.freeservers.com/

3. An Introduction to Support Vector Machineshttp://www.support-vector.net/

4. Archives of SUPPORT-VECTOR-MACHINES ...http://www.jiscmail.ac.uk/lists/SUPPORT...

5. SVM-Light Support Vector Machine http://ais.gmd.de/~thorsten/svm light/

1. Kernel Machines http://svm.first.gmd.de/

2. SVM-Light Support Vector Machine http://ais.gmd.de/~thorsten/svm light/

3. Support Vector Machine and Kernel ... Referenceshttp://svm.research.bell-labs.com/SVMrefs.html

4. Lucent Technologies: SVM demo applet http://svm.research.bell-labs.com/SVT/SVMsvt.html

5. Royal Holloway Support Vector Machine http://svm.dcs.rhbnc.ac.uk

1. Kernel Machines T2http://svm.first.gmd.de/

2. Support Vector Machine T1http://jbolivar.freeservers.com/

3. SVM-Light Support Vector Machine T2http://ais.gmd.de/~thorsten/svm light/

4. An Introduction to Support Vector Machines T1http://www.support-vector.net/

5. Support Vector Machine and Kernel ... References T2http://svm.research.bell-labs.com/SVMrefs.html

6. Archives of SUPPORT-VECTOR-MACHINES ... T1http://www.jiscmail.ac.uk/lists/SUPPORT...

7. Lucent Technologies: SVM demo applet T2http://svm.research.bell-labs.com/SVT/SVMsvt.html

f1(u,q) r1 f2(u,q) r2

Interleaving(r1,r2)

(u=thorsten, q=“svm”)

Interpretation: (r2 Â r1) ↔ clicks(T2) > clicks(T1)

Invariant: For all k, in expectation same number of team members in top k from each team.

NEXTPICK

[Radlinski, Kurup, Joachims; CIKM 2008]

Page 5: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Dueling Bandits Problem

• Continuous space bandits F – E.g., parameter space of retrieval functions (i.e., weight vectors)

• Each time step compares two bandits– E.g., interleaving test on two retrieval functions– Comparison is noisy & independent

Page 6: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Dueling Bandits Problem

• Continuous space bandits F – E.g., parameter space of retrieval functions (i.e., weight vectors)

• Each time step compares two bandits– E.g., interleaving test on two retrieval functions– Comparison is noisy & independent

• Choose pair (ft, ft’) to minimize regret:

• (% users who prefer best bandit over chosen ones)

T

tttT ffPffP

1

1)'*()*(

Page 7: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

T

tttT ffPffP

1

1)'*()*(

•Example 1•P(f* > f) = 0.9•P(f* > f’) = 0.8•Incurred Regret = 0.7

•Example 2 •P(f* > f) = 0.7•P(f* > f’) = 0.6•Incurred Regret = 0.3

•Example 3•P(f* > f) = 0.51•P(f* > f) = 0.55•Incurred Regret = 0.06

Page 8: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Modeling Assumptions

• Each bandit f 2F has intrinsic value v(f)– Never observed directly– Assume v(f) is strictly concave ( unique f* )

• Comparisons based on v(f)– P(f > f’) = σ( v(f) – v(f’) )– P is L-Lipschitz

– For example: )exp(1

1)(

xx

Page 9: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Probability Functions

Page 10: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Dueling Bandit Gradient Descent

• Maintain ft

– Compare with ft’ (close to ft -- defined by step size)

– Update if ft’ wins comparison

• Expectation of update close to gradient of P(ft > f’)– Builds on Bandit Gradient Descent [Flaxman et al., 2005]

Page 11: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

δ – explore step size γ – exploit step sizeCurrent pointLosing candidateWinning candidate

Dueling Bandit Gradient Descent

Page 12: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

δ – explore step size γ – exploit step sizeCurrent pointLosing candidateWinning candidate

Dueling Bandit Gradient Descent

Page 13: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

δ – explore step size γ – exploit step sizeCurrent pointLosing candidateWinning candidate

Dueling Bandit Gradient Descent

Page 14: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

δ – explore step size γ – exploit step sizeCurrent pointLosing candidateWinning candidate

Dueling Bandit Gradient Descent

Page 15: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

δ – explore step size γ – exploit step sizeCurrent pointLosing candidateWinning candidate

Dueling Bandit Gradient Descent

Page 16: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

δ – explore step size γ – exploit step sizeCurrent pointLosing candidateWinning candidate

Dueling Bandit Gradient Descent

Page 17: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

δ – explore step size γ – exploit step sizeCurrent pointLosing candidateWinning candidate

Dueling Bandit Gradient Descent

Page 18: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

δ – explore step size γ – exploit step sizeCurrent pointLosing candidateWinning candidate

Dueling Bandit Gradient Descent

Page 19: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

δ – explore step size γ – exploit step sizeCurrent pointLosing candidateWinning candidate

Dueling Bandit Gradient Descent

Page 20: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Analysis (Sketch)

• Dueling Bandit Gradient Descent– Sequence of partially convex functions ct(f) = P(ft > f)

– Random binary updates (expectation close to gradient)

• Bandit Gradient Descent [Flaxman et al., SODA 2005]

– Sequence of convex functions – Use randomized update

(expectation close to gradient)

– Can be extended to our setting

(Assumes more information)

Page 21: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Analysis (Sketch)

• Convex functions satisfy

– Both additive and multiplicative error– Depends on exploration step size δ – Main analytical contribution: bounding multiplicative error

*)()(*)()( xxxcxcxc

Page 22: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Regret Bound

• Regret grows as O(T3/4):

• Average regret shrinks as O(T-1/4)– In the limit, we do as well as knowing f* in

hindsight

T

tttT ffPffP

1

1)'*()*(

RdLTT 102E 4/3

δ = O(1/T-1/4 )γ = O(1/T-1/2 )

Page 23: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Practical Considerations

• Need to set step size parameters– Depends on P(f > f’)

• Cannot be set optimally– We don’t know the specifics of P(f > f’)– Algorithm should be robust to parameter settings

• Set parameters approximately in experiments

Page 24: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

00.10.20.30.40.50.60.70.80.9

1

10 570

1130

1690

2250

2810

3370

3930

4490

5050

5610

6170

6730

7290

7850

8410

8970

9530

Aver

age

Regr

et

Regret Comparison DBGD vs BGD

DBGD

BGD 1

BGD 2

• 50 dimensional parameter space• Value function v(x) = -xTx• Logistic transfer function• Random point has regret almost 1

More experiments in paper.

Page 25: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Web Search Simulation

• Leverage web search dataset– 1000 Training Queries, 367 Dimensions

• Simulate “users” issuing queries– Value function based on NDCG@10 (ranking measure)– Use logistic to make probabilistic comparisons

• Use linear ranking function.

• Not intended to compete with supervised learning– Feasibility check for online learning w/ users– Supervised labels difficult to acquire “in the wild”

Page 26: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

• Chose parameters with best final performance• Curves basically identical for validation and test sets (no over-fitting)• Sampling multiple queries makes no difference

0.480.5

0.520.540.560.58

0.60.62

063

0000

1260

000

1890

000

2520

000

3150

000

3780

000

4410

000

5040

000

5670

000

6300

000

6930

000

7560

000

8190

000

8820

000

9450

000

Trai

ning

NDC

G @

10Web Simulation Results

Sample 1

Sample 10

Sample 100

Ranking SVM

Page 27: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

What Next?

• Better simulation environments– More realistic user modeling assumptions

• DBGD simple and extensible – Incorporate pairwise document preferences– Deal with ranking discontinuities

• Test on real search systems– Varying scales of user communities– Sheds on insight / guides future development

Page 28: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Extra Slides

Page 29: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Active vs Passive Learning

• Passive Data Collection (offline)– Biased by current retrieval function

• Point-wise Evaluation– Design retrieval function offline– Evaluate online

• Active Learning (online)– Automatically propose new rankings to evaluate– Our approach

Page 30: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Relative vs Absolute Metrics

• Our framework based on relative metrics– E.g., comparing pairs of results or rankings– Relatively recent development

• Absolute Metrics– E.g., absolute click-through rate– More common in literature – Suffers from presentation bias– Less robust to the many different sources of noise

Page 31: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

What Results do Users View/Click?

[Joachims et al., TOIS 2007]

Page 32: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem
Page 33: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Analysis (Sketch)

• Convex functions satisfy

– We have both multiplicative and additive error– Depends on exploration step size δ – Main technical contribution: bounding multiplicative error

*)()(*)()( xxxcxcxc

T

tttt ffPffP

1

*)()(E

Existing results yields sub-linear bounds on:

Page 34: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Analysis (Sketch)

• We know how to bound

• Regret:

• We can show using Lipschitz and symmetry of σ:

T

tttT ffPffP

1

1)'*()*(

LTffPffPT

ttttT

1

*)()(E2E

T

tttt ffPffP

1

*)()(E

Page 35: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

More Simulation Experiments

• Logistic transfer function σ(x) = 1/(1+exp(-x))• 4 choices of value functions

• δ, γ set approximately

Page 36: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

TR

Page 37: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

NDCG• Normalized Discounted Cumulative Gain• Multiple Levels of Relevance

• DCG:– contribution of ith rank position:

– Ex: has DCG score of

• NDCG is normalized DCG – best possible ranking as score NDCG = 1

)1log(

12

i

iy

45.5)6log(

1

)5log(

0

)4log(

1

)3log(

3

)2log(

1

Page 38: Interactively Optimizing  Information Retrieval Systems as  a Dueling Bandits Problem

Considerations

• NDCG is discontinuous w.r.t. function parameters– Try larger values of δ, γ– Try sampling multiple queries per update

• Homogenous user values– NDCG@10– Not an optimization concern– Modeling limitation

• Not intended to compete with supervised learning– Sanity check of feasibility for online learning w/ users