Polemic tweet - IFIP Interact 2013

PolemicTweet
Video Annotation and Analysis
through tagged Tweets
Samuel Huron, Petra Isenberg, Jean Daniel Fekete
06/09/13 IFIP Interact 2013 – Polemic tweet

.

Petra
Isenberg
INRIA
@dr_pi
Raphael
Velt
IRI
@raphv
Yves Marie
Haussonne
IRI
@ymh_work
Authors and Contributors
Samuel
Huron
INRIA - IRI
@cybunk
Jean Daniel
Fekete
INRIA
@jdfaviz

At IRI we organize seminars
on subjects such as
philosophy, design, and
digital cultures. !

SITUATION

Previous process
1. Videotape
the seminar
2. Manually
annotate and
tag
3. Publish the
video online and
let user annotate

PROBLEM

Problem
2. Annotating a
video manually
is boring and
time-consuming
1. Browsing a
video after a
conference is
difficult
3. Tagging and
structuring
annotations are
even more
boring

Inspiration: Microblogging activities
Communication devices

Motivation
3. Provide an easily
accessible system
from annotation to
video analysis.
2. Provide a visual
backchannel to
incite people to
tag their tweets.
1. Engage the
audience:
crowdsource video
annotation and
tagging by tweet.

A certain type of tag: Sentiments

Current approaches to extract
sentiments
Human analysis
Natural Language
Processing

Our need
  Fast
  Language agnostic
  Accurate
  Cheap

Why not ask the tweet autor ?
Cost Benefit
- Fast
- Language agnostic
- Accurate
- Cheap, cost is distributed
- Require user effort
- Low recall

DESIGN

Design Goals:
1. To crowdsource
conference video
tagging with
structured
sentiment metadata.
2. To engage
audiences in the
tagging process.
3. To visualize
annotations for
browsing and
analyzing the video.

0. Service Design
1. A special syntax to
enhance tweets
4. A video player with a
visualization
2. A social protocol to
organize the event over time
3. A real time visualization to
read and write tweets

1. Syntax properties and description:
GreenAgreement++
RedDisagreement--
YellowReferences==
BlueQuestions??
Tag Semantic Color
  Simple
  Brief
  Neutral
  Easy to parse
  Expressive
  Unambiguous
Design Rational

2. Social Protocol
A. Before the
talk: Instruct
the crowd.
B. During the talk:
Crowdsource the
video tagging with
a backchannel.
C. After the
talk: Publish
on a web video
player.

Instruct the crowd
  The network
access policy
  The Twitter
hashtag
  The website
URL
  The tags

3. Backchannel: Real Time Visualization

4. Metadata Video Player
http://goo.gl/zyJYCZ

DEPLOYMENT

25 to
Attendees
per event
27
Events
9088
Tweets
recorded
300 1012
Unique
Twitter
users
Deployment

EVALUATION

Questions
1. Is the system
sufficiently
engaging?
3. What is the
impact of
presenting tweets
and tags on a web
video player?
2. Does the visual
backchannel
incite people to
add tags?

Data Collection
Record tweets sent
during the events
Log the user
interaction on the
web video player
Two web surveys,
one for
participants, one
for organizers

RESULT
IFIP Interact 2013 – Polemic tweet05/09/2013

Question 1:
Is the system sufficiently engaging?
0.2 to 7.94Tweets
per minute
0.12 to 2.13 Tags
per minute
40% Tweets using
our syntax
35 to 45 % Of all individuals
used our syntaxe

Question 1:
Is the system sufficiently engaging?
•  Type of Audience. (size, culture, equipment, incentive, distribution)
•  Type of Speaker and Twitter Reaction.
•  Type of Content.
•  Observance of the PT Protocol.
Influential factors for engagement

Question 2:
Does the visual backchannel
incite people to add tags?

Question 3:
What is the impact of presenting tweets and
tags on a web video player?

Question 3:
Visualization User seek activity
?

Question 3:
TODO: Figuresp < 0.1 p < 0.1 p < 0.1
0.1 < p < 0.5 0.1 < p < 0.5P > 0.1

CONCLUSION

The Evaluation was challenging
  Deployed in the wild
  Long period of time
  Design change over time
  Real life users
  Real life setting
Because

Result!
1.The system
engaged the
audience to provide
data for
crowdsourced video
tagging.
2. The visual
backchannel incited
people to tag their
tweets and thus
provided an
annotation
structure.
3. The media player
augmented with the
annotated tweet
visualization was
considered a
significant
improvement over
traditional video
players.

Next steps
As classroom
presentations.
Synchronous
web seminar.
Public debates,
and social TV.

Possible optimizations:
Shared attention. Tag properties. Scalability, user
interface design and
system.

Subsequent work:
http://goo.gl/c2iqg0
Bubble-TV: Live Visual
Feedback for Social
TV Broadcast
Visual Sedimentation:
Data stream
visualization technique
http://goo.gl/KvllnO
User supplied
sentiment in tweets
http://goo.gl/XXVzJb

MORE INFOS:
+ open source
+ details
THANKS: To people who used the system and people who fund this work:
Thank you !
Contact :
Samuel.Huron@inria.fr
http://www.cybunk.com
Team :
http://www.aviz.fr
http://iri.centrepompidou.fr
goo.gl/wKG9Sg

THE END

Survey Result
  96% (25) of respondents attended at least one
conference had already used twitter at conferences
( 7% only for reading)
  86% (24) reported familiarity with the PT syntax
  21% (6) with the LiveTweet interface
  35% (14) with the web video player visualization
timeline.
05/09/2013 WIID: Who Is Interaction Design?

Survey Result
In the survey 14 users reported to know
what the Tweet Timeline Visualization is.
05/09/2013 WIID: Who Is Interaction Design?
• 64% (9) of these found that
provides useful information
• 79% (11) agree or strongly agree
that the visualization helps them to
browse the video
• 86% (12) would recommend PT to
a friend.
# to view a part of video with :
– many tweets: 57% (8)
– positive opinions (++): 29% (4)
– negative opinions (–): 43% (6)
– questions (??): 21% (3)
– references (==): 21% (3)

Cost and benefit of tagging technics
- Time consuming
- Cost is proportional
- Hight recall (100%)
- Mostly targeted at English
- Not “off the shelves”
- Precision problem
- Cost is proportional
- Language agnostic
- Accurate
- Fast
Cost Benefit
Mechanical turk
- Language agnostic
- Accurate
- Fast
- Cheap, cost is distributed
- Engage user
- Low recall
Natural Language
Processing

Polemic tweet - IFIP Interact 2013

More Related Content

Similar to Polemic tweet - IFIP Interact 2013 (11)

Recently uploaded (20)

Polemic tweet - IFIP Interact 2013