Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti...

25
Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute of Technology, Bombay

Transcript of Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti...

Page 1: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Learning Conference Reviewer Assignments

Adith Swaminathan

Guide :

Prof. Soumen Chakrabarti

Department of Computer Science andEngineering,Indian Institute of Technology, Bombay

Page 2: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Future Work (from BTP1)

Given WWW2010’s assignments, learn Affinity_Param, Topic_Param and Irritation

Citations as edge featuresLoad-Constrained Partial AssignmentsBetter estimation of Assignment Quality

Page 3: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Background

Conference Reviewer-Paper Assignment as a Many-Many-matching [1]

Minimum Cost Network Flow (MCF)

Page 4: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Conference Reviewer Assignment

Set of Reviewers, R, max #papers = L_i Set of Papers, P, min #reviews = K

Assumption : Only require #reviews, not quality

Suppose we have cost function A_ij(y) for <R_i, P_j>

Page 5: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

ILP -> Assumption -> MCF

Page 6: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Two problems

Integer Linear Programs are NP-Hard!– Relax?– More assumptions?

How to determine A_ij?– M * N ~ 10000– Multimodal clues

Page 7: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

ILP -> Assumption -> MCF

Enforce structure on A_ij– Better model multimodality– Fewer parameters to fix

“Learn” A_ij using Structured Learning Techniques

A_ij = wT Φ(R_i, P_j, y_ij)

Page 8: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Ramifications of Structured Costs

Costs decompose over <R_i, P_j> pairs– Decomposable Preference Auction– Polynomial Algorithms for DPAs [2]

Restricted notion of optimality– Per-reviewer/Per-paper constraint could be

combinatorial– Stability?

Page 9: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

ILP -> Assumption -> MCF

Page 10: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Minimum Cost Network Flow

Directed graph G=(V,E), capacities u(E)>= 0, costs c(E)

Nodes have numbers b(V) : Sum(b(V)) = 0

Task : Find a function f: E->R+ which satisfies the b-flow at minimum cost

Successive Shortest Path Algorithm

Page 11: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Node features and Edge features

Profile

Topics

Reviewer

Topics

Contents

Paper

Affinity

Bid

Topic Overlap

Cites

Page 12: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

The Loss Function

L_ij = w_1 * exp(-Affinity_ij) + w_2 * [[1 – Topic_Overlap_ij]] + w_3 * Bid_Cost

Bid_Cost = Potential(R_i, P_j, y_ij)

Irritation (I) and Disappointment (D) needs to be set

Page 13: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Assignment Quality Measures

Number of Bids Violated?– Not a reliable measure.

+ve Bids Violated–ve Bids ViolatedAssignments satisfying Topic MatchConfidence?

Page 14: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Confidence == Quality?

Very sparse– Fewer than 5% observed– Extrapolated Confidence?

Reliable– Bids as a precursor of Confidence [3]

– Confidence-Augmented Loss?

Page 15: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Learning w’s

Transductive Ordinal Regression– Assume : Assignments are independent (Naïve)– Heuristic : Augment observed dataset– Extrapolate observed Confidence [4]

– Learn w over extrapolated dataset

Support Vector Machine for Structured Outputs– Cast as soft-margin SVM formulation [5]

– Upper-bound objective with a convex fn (Optimality?)– Minimize, using Cutting Plane (Approximate)

Page 16: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Transductive Ordinal Regression [6]

Page 17: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

SVM Struct. [7]

Loss Augmented Inference ~ Most Violated ConstraintLoss is decomposable -> Modified MCF

Page 18: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

PA

RA

: P

aper

Ass

ign

men

t to

R

evie

wer

s A

pp

arat

us

Page 19: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Results

Page 20: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Bimodal Behaviour

•Reviewer either gets few or L_i papers•Load Penalties [8]

•Introduce more parameters•Infer using modified MCF•Learning parameters?

•Load Rebalancing•Tradeoff between MCF optimum and old assignment

Page 21: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Penalise Reviewer Loads

Page 22: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Load Constrained Assignments

Page 23: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

Avenues for Future Work

•Document Modelling for Affinity Scores

•Objective Assignment Evaluation

•Transitive Citation Scores

•Load Penalty Parameter Estimation

Page 24: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

References

1. The Conference Paper Assignment Problem, J. Goldsmith, R.H. Sloan, 2007

2. MultiAgent Systems: Algorithmic, Game-Theoretic, and Logical Foundations, Y. Shoham, K. Leyton-Brown, 2009

3. Automating the Assignment of Submitted Manuscripts to Reviewers, S.T. Dumais, J. Nielson, 1992

4. Semisupervised Regression with cotraining algorithms, Z. Zhou, M. Li, 2007

Page 25: Learning Conference Reviewer Assignments Adith Swaminathan Guide : Prof. Soumen Chakrabarti Department of Computer Science and Engineering, Indian Institute.

References – contd.

5. Learning structured prediction models : A Large Margin Approach, B. Taskar, et al, 2005

6. Ologit : Ordinal Logistic Regression for Zelig, G. King, et al, 2007

7. SVM Learning for Interdependant and Structured Output Spaces, I. Tsochantaridis, et al, 2004

8. Word Alignment via Quadratic Assignment, S. Lacoste-Julien, et al, 2006