How do people tag pictures?courses.cecs.anu.edu.au/courses/CS_PROJECTS/11S1/Final present… ·...
Transcript of How do people tag pictures?courses.cecs.anu.edu.au/courses/CS_PROJECTS/11S1/Final present… ·...
1
How do people tag pictures?- A Pilot Study with Facebook Application
Student: Zhe QinSupervisor: Lexing XieJune 9th, 2011
ANU College of Engineering & Computer Science
2
ANU College of Engineering & Computer Science
Flickr – Source of Photos and Tags
3
ANU College of Engineering & Computer Science
Flickr – Popular Tags
The size represents the popularity of the tag
4
ANU College of Engineering & Computer Science
GWAP: Game With A Purpose
Developed by Luis von Ahn and Laura Dabbish of Carnegie Mellon University in 2003.
The idea: human brainpower through computer games VS artificial intelligence algorithms.
Collected more than 10 million image labels in few months.
Examples in photo tagging: ESP Game, Google Image Labeler.
“People play not because they are personally interested in solving an instance of a computational problem but because they wish to be entertained.”
5
ANU College of Engineering & Computer Science
GWAP: ESP Game
.
1 6
5
4
3
Design principles:① High-score list② Player skill levels③ Score keeping④ Taboo words⑤ Timed response⑥ Score bonus
2
6
ANU College of Engineering & Computer Science
Objectives:
Figure out which tags are visually recognizable and the difference between author’s tags and view’s tags for the same photos.
The work:
Teamwork of developing a proper application (a Facebook application, amobile application or an Amazon Mechanical Turk) as a tool to collect user tagsof photos.
Team members: Li Li and me.
Our Project
7
ANU College of Engineering & Computer Science
The Application – Tag Expert
My part Li Li’s part
8
ANU College of Engineering & Computer Science
Choosing the Platform – FacebookThe Amazon Mechanical Turk (MTurk)
•Limited number of workers
•Cost money if want efficiency
Mobile phone platforms
•Large number of potential users
•Almost accessible everywhere
•No guarantee of real users
•Most popular social networking website
•Large number of active users
•Easy to promote by friends
9
ANU College of Engineering & Computer Science
My Work – Web Application
The web application consists of four main functions:Select photos sets on Flickr and send them with tags to Facebook friends for the purpose of tagging.
Tag photos sent by friends, and be able to view the original tags afterwards.
View feedback (tags) of sent photos from friends.
Invite their Facebook friends to use “Tag Expert”.
10
ANU College of Engineering & Computer Science
My Work – Web Application
Users go to here first
11
ANU College of Engineering & Computer Science
My Work – Facebook Integration
The idea: Facebook load and display web apps within an iframe in the standard Facebook chrome.
Two ways of doing this: FBML Canvas and Iframe Canvas.
Tag Expert
Canvas Page
12
ANU College of Engineering & Computer Science
My Work – Facebook Integration
FBML
IFrame
Our Choice
Depreciated by Facebook
13
ANU College of Engineering & Computer Science
My Work – How It Works
HomepageNavigation Bar to
•Send page•View tags page•About page
Go to tag pageGo to invite page
14
ANU College of Engineering & Computer Science
My Work – How It Works
Tag page
Input tags separated by “space”
15
ANU College of Engineering & Computer Science
My Work – How It Works
Send pageSelect one photo set and
multiple friends
1
2
3Assign tags for each photo sent (Similar to Tag Page)
16
ANU College of Engineering & Computer Science
My Work – How It Works
View tags page
17
ANU College of Engineering & Computer Science
My Work – How It Works
Invite page
Select friends, and a request notice will display on their
Facebook Homepage
18
ANU College of Engineering & Computer Science
My Work – How It Works
About page
Information Sheet. Required by ANU Human Research Ethics Committee (HREC)
19
ANU College of Engineering & Computer Science
My Work – Data Analysis
Due to the limited time, we are only able to do one of the Scenarios we designed within the development team.
Scenario 1:We design three versions of our app interface: one is the original; in the secondversion, tagging for each photo will be required to finish within 10 seconds or itwill pass on to next photo while the countdown is shown to the user; based onthe second version, the third version enables a score keeping if each tagging isfinished in time and the score will be display to the user. Each version isassigned with a different photo set. Then we ask every participant to play allthree versions.
To test if timed response and score keeping could effect the correctness and efficiency of data collection
20
ANU College of Engineering & Computer Science
My Work – Data Analysis
Number of tags Participant 1 Participant 2 Participant 3Version 1 50 60 54
Version 2 30 35 40
Version 3 45 40 47
Time used Participant 1 Participant 2 Participant 3
Version 1 294 seconds 385 seconds 370 seconds
Version 2 176 seconds 171 seconds 185 seconds
Version 3 194 seconds 196 seconds 192 seconds
Matching ratio Participant 1 Participant 2 Participant 3
Version 1 41% 35% 52%
Version 2 44% 20% 47%
Version 3 56% 50% 67%
Time response could possibly reduce the time of collection data while not affecting the correctness of data collected.
Score keeping with time response could improve both the correctness and efficiency.
21
ANU College of Engineering & Computer Science
Future Work
A
B
Possibly to figure out the type of the picture