KissKissBan: A Competitive Human Computation Game for Image Annotation

Post on 18-Dec-2014

1,056 views 2 download

description

In this paper, we propose a competitive human computation game, KissKissBan (KKB), for image annotation. KKB is different from other human computation games since it integrates both collaborative and competitive elements in the game design. In a KKB game, one player, the blocker, competes with the other two collaborative players, the couples; while the couples try to find consensual descriptions about an image, the blocker’s mission is to prevent the couples from reaching consensus. Because of its design, KKB possesses two nice properties over the traditional human computation game. First, since the blocker is encouraged to stop the couples from reaching consensual descriptions, he will try to detect and prevent coalition between the couples; therefore, these efforts naturally form a player-level cheating-proof mechanism. Second, to evade the restrictions set by the blocker, the couples would endeavor to bring up a more diverse set of image annotations. Experiments hosted on Amazon Mechanical Turk and a gameplay survey involving 17 participants have shown that KKB is a fun and efficient game for collecting diverse image annotations.

transcript

KissKissBanA Competitive Human Computation

Game for Image Annotation

Chien-Ju Ho*, Tao-Hsuan Chang+, Jong-Chuan Lee+, Jane Yun-jen Hsu+, Kuan-Ta Chen*

*Academia Sinica+National Taiwan University

Human Computation Game

Collaborative Mechanisms

Coalition Problem

Limited Data Diversity

The ESP Game

• Coalition problem

• partner with themselves

• unified strategy

• Limited diversity

• easy words first

Competitive Mechanism?

DOGFLOWERFLOOR

CUTEWOODDOG

Player 1 Player 2

DOGFLOWERFLOOR

CUTEWOODDOG

DELIVER EMOTION BY TYPING THE SAME WORD

Player 1 Player 2

DOGFLOWERFLOOR

CUTEWOODDOG

DELIVER EMOTION BY TYPING THE SAME WORD

Couple Couple

DOGFLOWERFLOOR

CUTEWOODDOG

DOGWHITE

Couple CoupleBlocker

DOGFLOWERFLOOR

CUTEWOODDOG

DOGWHITE

Couple CoupleBlocker

KissKissBanA Competitive Human Computation

Game for Image Annotation

Couple CoupleBlocker

1. First, the blocker has 7 seconds to type words

DOGWHITE

The blocked words

2. Then, the couples has 30 seconds to type words

DOGFLOWERFLOOR

CUTEWOODFLOWER

-5s game time -5s game time

Blocked

Couple CoupleBlocker

1. First, the blocker has 7 seconds to type words

DOGWHITE

2. Then, the couples has 30 seconds to type words

DOGFLOWERFLOOR

CUTEWOODFLOWER

Couples win for matching within time limit

Screenshots of KissKissBan

Intuitions Behind the Game

Incentive Structure

• Zero sum game

• strict competitive

• players can not benefit from coalition

• Different decision time periods

• harder words, longer consideration [Barry et al. 2001]

• diverse data contribution

Small-scale Experiment

• Image Source

• 125 images randomly from ESP dataset

• Experiment Setup

• 17-player gameplay survey

• Data collection on Mechanical Turk

Gameplay Survey

• Fun

• 3.76/5 overall rating

• 15/17 players claim that would play again

• over 60% like to play as a blocker

MTurk Experiment

• Collecting data from anonymous users

• Publish two games for comparison

• KissKissBan

• ESP Game without taboo words

• Each game is played for ~5000 rounds

Example of Collected DataESP ML-KKB BL-KKB

man 21beach 10karate 5water 1

beach 3water 3sand 3sea 2

ninja 1kungfu 1ocean 1

sea 9man 8

ocean 3black 1china 1sand 1

ML-KKB: matching between couplesBL-KKB: matching between couple and blocker

MTurk Experiment

• Data property

• For each image

• ESP: 39.96 labels, 6.56 distinct labels

• KKB: 44.17 labels, 11.53 distinct labels

• A more diverse player contribution

Conclusion

• We proposed a competitive human computation game, KissKissBan.

• KissKissBan has two nice properties

• a player-level anti-coalition mechanism

• diverse player input

More in the Future

• Publish the game on Facebook within the next one to two months.

• Study the different tagging behaviors of the blocker and the couples.

• Apply the competitive mechanism to more general human computation game design.

Q & A