Revised version for the CBS Seminar with further findings in Sentiment Analysis leveraging automatic systems from Dan Hardt, manual coding from corporate communications students, and input Bernardo Huberman of HP Labs.
4. Overview
1.
2.
3.
4.
5.
6.
The Story: Timeline of Online/Offline Events
Dataset, Tools and Visualization
Research Angles
Volume Comparison
Community Language & Demographics
Social Graph: Network, Actor Post-Level
Analysis
7. Social Text: Language & Sentiment Analysis
8. Company Performance
9. Initial Findings
6. Research Avenues of Inquiry
The online reflection - Why does this matter?
• Volume – How did the conversation amplitude evolve over
two weeks in February online?
• Sentiment – Where did negative sentiment originate and how
did it evolve/spread? (keywords, people, and topics)
• Community – Who were the relevant actors?
(Organizations, Customers, Users, Activists, Influencers, etc.)
• Post-level Performance – What types of posts and specific
events instigated the issue online? (artifacts involved such as
videos, photos, Facebook posts, tweets, etc)
How did the Copenhagen Zoo handle the event on social
channels and how did the (social) media storm effect their
presence? How did other organizations deal with the crisis?
What made this incident different and how?
7. Toolset
Tool
Purpose
Access
Radian6
Collection
License
Nitrogram
Collection
License
Tableau Desktop Visualization / Analysis
License (edu)
Datawrapper
Visualization
Public
TimelineJS
Visualization
Public
LIWC
Language Analysis
Public
SODATO
Collection / Vizualization Beta
Topsy Pro
Collection / Analysis
Trial
Scoailbakers
Facebook Statistics
COTS*
Followerwonk
Context
COTS*
Twtrland
Context
COTS*
Quintly
Context
COTS*
Wildfire
Historical Performance
COTS*
Consumer of the shelf tool (COTS)
8. #Marius Overview
Social Data Collected
• 40 Online Channels (Jan 19 – Feb 19)
• Over 315 K Posts Collected (75% Twitter)
• 200 K Unique Posts (63%)
• 681 Million Potential Impressions on Twitter
• Highest Buzz Rate : 332 Posts / Minute
Normal Monthly Volume : 300-500 Stories
• 30K Petition Signatures
• 45K Facebook Protesters
13. Channel Comparison
•
•
Twitter dominates 75% of total
chatter, while 21% is from Facebook
Discussions
Amplification: 50% of Tweets are
retweets
Danish Subset
• Media channels are more rich in
diversity
• Facebook and Twitter only share
half the conversation
• Only a quarter of all Tweets are retweets
> Does mainstream media play a
greater role for Danish society
while, social media is dominant
elsewhere in terms of quantity of
discussion and breadth of dispersion?
14. Region and Language
Detection
• 95% of the total
conversation was
detected to be in English.
• Almost two thirds of
global activity came from
the US (64%), followed by
the UK (13%) and
Netherlands (4%).
• Danish was only detected
in 2,220 posts.
15. #Marius Demographics
Location estimates
North America
usage over 50%
Twitter bio field
reveals several
dominant traits
during the weekend:
•
•
•
•
•
•
Liberal,
Progressivism,
Vegan,
Activist,
Animal rights,
advocate, pets, wil
dlife, etc
24. Automatic Sentiment Results
• Danish data tends to be
much more neutral
compared to the nonDanish data.
• Most of the negativity
detected in Twitter for
non-Danish data while
most of the negative
data occurs in
Facebook for Danish
data.
> Does this imply that
Danes prefer Facebook to
Twitter to express their
ideas?
27. Copenhagen Zoo Facebook
Performance
• Largest surge in likes ever
• Almost 100K People
Talking About This (PTAT)
on Facebook
• 120.3% Normalized Buzz
(PTAT/Likes)
Global Fan Growth
• Over 10K new fans this
month (70% in Denmark)
• 19 Countries more than
doubled their fanbase
• Countries such as the UK
and Australia tripled and
almost quadrupled their
fanbases of CPH Zoo.
Likes
PTAT
PTAT / Likes
29. Check-ins
• Beforehand, 29K
people added the
Zoo’s location to a
Facebook post
• Now 110K people
“Were Here” on
Facebook
• CPH Zoo is thus now
the 7th most checkedinto place in Denmark
30. Initial Findings
Overall
• Twitter offered a more direct reflection of events, in terms
of volume and sentiment
• Twitter also demonstrated a more drastic reaction to
network prestige factors from activists and celebreties
• Automatic sentiment on Radian6 is neutral-heavy, often
failing to detect negative sentiment,
• The dominance of English-language countries and the
Twitter channel went hand-in-hand (perhaps along with
mainstream spin).
• The mechanisms on Facebook allow a dichotomy from
crisis situations by yielding negative sentiment in terms of
comments and posts, while simultaneously experiencing
unprecedented growth in positive signals (such as fans
and likes, as well as buzz and check-ins).
31. Danish Comparison
Contrast with Danish Subset
• Mainstream media plays a larger role as
opposed to higher proportions of online
debate on social channels elsewhere
• Re-tweets levels are relatively small and
social media may be used social media
more to express oneself rather than to
share information.
• Negative sentiment was detected more
strongly on Facebook
Legacy of this Giraffe - Work in progressNext thing – another animal rights incident with a whale has come and gone, twitter has long since moved onNonetheless still tell the story of an explosion on social media with some pretty rich and large data
- For both Danish and non-danish the major media providers are Twitter and Facebook, but the media channel is more diverse for non-danish data, without the predominance of Twitter- Given that retweet take a more important role for non-danish data while Facebook is more popular for Danish, does mainstream media play a greater role for Danish society while in non-danish data, social media is now dominant in terms of quantity of discussion and breadth of dispersion?- It is interesting to see that Retweets for Danish data are proportionally small. Does this imply that Danes use social media more to express themselves rather than to share information?
People with large followers also play an essential role in spreading the information. For example @rickygervais ( English comedian) and @fielaursendk( Danish blogger)
Sources : Sentiment140, Twtrland
Lets talk about Facebook, after the twitter-heaviness of the overall online conversation