IBM System G Social Media Solution Tutorial

IBM System G Social Media Solution Tutorial
Principal Investigator: Ching-Yung Lin
!
Social, Cognitive and Network Science Group
IBM T. J. Watson Research Center
April 20, 2015
© 2015 IBM Corporation
Outline
•
•
•
•
•
•
•
•
•
•
•
•
•
2
Registration
Live Monitoring
Trend Monitoring
Multimedia Monitoring
Geo Monitoring
Scope Identification
Segment Analysis
Link Exploration
Impact Prediction
Story Detection
Person Analytics
Target Discovery
Anomaly Detection
© 2015 IBM Corporation
Registration
Link: https://systemg.mybluemix.net/solution/socialmedia/
create IBM id to sign in
3
© 2015 IBM Corporation
Registration — Agreement
4
© 2015 IBM Corporation
Overview Page
5
© 2015 IBM Corporation
Live Monitoring
Monitor live tweets and the retweet graph on a channel or
given an input keyword
• Data is refreshed every 3 seconds. If the user does not explicitly STOP the
monitoring, the monitoring shuts down after 5 minutes (elapsed time).
tweet
statistics
(sentimen
ts are
detected
from
tweet
text)
system status
geo-locations of
the tweets
popular
hash tags
retweets
original
tweets
6
retweet graph
(each node is one account and each edge is a
tweet that got retweeted by another account)
© 2015 IBM Corporation
Live Monitoring — Setup
After setting up a channel(either by selecting from drop down menu or inputting keywords), the
live tweet from this time point will be showing on the screen.
Method 2: type in keywords to
filter the tweets and click on
“GO”
Method 1: select a channel from the
drop down menu
7
© 2015 IBM Corporation
Live Monitoring — Usage
hover over the edge to show the
tweet that one node retweeted
from the other node
Each entry shows person
sending the original tweet with
(# of followers), then an arrow
“->” showing the retweets. # of
retweets are shown at the end
of each entry
hover over the node to
show the twitter ID and
enlarged profile picture
“live” graph showing tweets and retweets. Size of
each node corresponds to the # of followers, i.e.
the bigger the node, the more followers it has.
Edge shows the people doing the retweets.
8
© 2015 IBM Corporation
Trend Monitoring
A time-line view of the popularity of topics/hashtags in a given channel
aggregation
methods of the
timeline view
color code of the
popular hashtags
popular
hashtags
(font size
represents
the
popularity)
raw
tweets
timeline view of the
percentage of
tweets containing
each hashtag
time of the day
(hour:minute:second)
9
© 2015 IBM Corporation
Trend Monitoring — Setup
Step 3: select
monitor interval
Step 1: select a channel from the
drop down menu or select “User
Input” to input filtering keywords
10
Step 2: select number of most popular hashtags to
be shown in the timeline from the drop down menu
or customize the number by selecting “User Input”
© 2015 IBM Corporation
Trend Monitoring — Usage
all previous tags: click on the dot of each hashtag to
show/hide tweets containing the hashtag from the timeline
Display (moving right to
left) is a time-lapsed
showing of the top x
hashtags for that topic.
Each new hashtag
generates a new color in
the display, but at each
time slice, only the top x
hashtags are shown.
click on the twitter
ID to see the
“Person Analysis”
of the twitter
account
hover over the time line for a given time-slice to see the
number of tweets containing each hashtag at each time
instance
11
© 2015 IBM Corporation
Multimedia Monitoring
Analyze the visual sentiment of live tweets which contains images in a channel
toggle to show only
original tweets or
retweets with
original tweets
twitter ID
adjacent-noun concepts
detected in the image
sentiment of the image
12
© 2015 IBM Corporation
Multimedia Monitoring — Setup
Method 2: type in keywords to
filter the tweets and click on
“GO”
Method 1: select a channel from the
drop down menu
13
© 2015 IBM Corporation
Geo Monitoring
Show live/past tweets in a selected channel/area on a map
click “Control Map” button to enable the function of changing map area
map zoom
in/out
tweet shown
on the map
with
correspondin
g physical
location
all tweets
that have
appeared
on the map
map area can be moved by drag and drop
14
© 2015 IBM Corporation
Geo Monitoring — Setup and Usage
Clicking on a given Twitter ID
(picture) will bring you to the
Person Analytics
Step 2: select past time period if viewing historical
tweets; skip this step is monitoring live tweets
Step 1: select a channel from the drop
down menu, input filtering keywords, or
input the name of a geo location
15
© 2015 IBM Corporation
Scope Identification
Scope identification helps users to define filter terms for a given channel
this star graph shows relationship (in terms
of occurrence) between terms in the channel
and the center term (isis in this example).
Scroll up and down to zoom in/out. Drag and
drop to rotate the globe.
16
© 2015 IBM Corporation
Scope Identification — Setup
select a channel from the drop down menu
17
© 2015 IBM Corporation
Scope Identification — Usage
click on a term to change the
relationship graph on the right
hover over to view the term and the cooccurrence between the term and the
center term.
click on the node will open a page on
Twitter showing all tweets containing the
corresponding hashtag
18
© 2015 IBM Corporation
Segment Analysis
Summarizes the distribution of Twitter user profiles in a channel
number of
users who
tweeted
particular
keywords
(clicking on
each
keyword
will filter the
tweets
using the
selected
keywords)
color code of the
map based on the
number of users in
each state
percentage of users in each category (gender, business owner, parents, or travelers)
19
© 2015 IBM Corporation
Segment Analysis — Setup and Usage
Step 1: select a channel from the drop
down menu
20
Usage: hover over a state to
see the number of user
profiles in the state
Step 2: Hovering over a given profile changes the color (e.g. from
Green to Orange) and changes the location indicated on the map
Clicking on more than one profile causes an AND of the conditions, the
count shows what’s left of the ones after the AND
© 2015 IBM Corporation
Link Exploration
Explore relations between Twitter users, tweets, hashtags, and time
relationship graph of the
tweets centering the hashtag/
screen name queried. The
nodes can be
a) hashtags
b) Tweeter user screen name
c) a tweet
d) time when the tweet is
posted
e) image in the tweet
21
© 2015 IBM Corporation
Link Exploration — Setup
Select a channel from the drop down
menu. The right hand graph will show
the initial layout
22
© 2015 IBM Corporation
Link Exploration — Usage
To explore the links, one can select the center of the graph by
•
click on any node in the graph
•
input query
The graph will re-layout centering the clicked node and showing all nodes connected to the
clicked node
23
© 2015 IBM Corporation
Link Exploration — Usage Example
First, query the IBM channel by hashtag “systemg”
step 1: select a query method
from the drop-down menu
step 2: input query and hit “Query”
graph centers at the node “Hashtag: systemg”.
Hover over the node to view details and all edges
connected to the node will be highlighted in orange
all tweets that contain hashtag system are
directly linked to the “Hashtag: systemg” node
time and images associated with the tweet
24
© 2015 IBM Corporation
Link Exploration — Usage Example
Then click any node to explore the links to the node. For instance, click on the “time” node of the
tweet “IBM System G team preparing Social Media Solution 2.0”:
another tweet that was posted at the same time
two Twitter users that participate in the tweet
25
© 2015 IBM Corporation
Impact Prediction
Predicting the impact of each tweet conversation
click to view the
impact features
keywords of the conversation, click to
view the conversation raw data
predicted
future trend
of impact
26
© 2015 IBM Corporation
Impact Prediction — Conversation Raw Data
Clicking on the keywords of a conversation, the raw data of the conversation is shown as below:
click on the profile picture to the twitter page of this tweet
27
© 2015 IBM Corporation
Impact Prediction — Features
Clicking on the “URL” of a conversation, the features for computing impact for this
conversation is shown as below. Feedback about the analysis result can also be inputed.
input feedback for the impact prediction
28
© 2015 IBM Corporation
Story Detection
Categorizing tweets that contains similar images into one story
images in tweets that belong to one story
text of the newest tweet in this story
toggle to show only
original tweets or
retweets with
original tweets
hover over an image to
show the text of the tweet
29
© 2015 IBM Corporation
Story Detection — Setup
Method 2: type in keywords to
filter the tweets and click on
“GO”
Method 1: select a channel
from the drop down menu
30
© 2015 IBM Corporation
Person Analysis
Analyze and aggregate the emotion and personal characteristics for a Twitter user profile over time
Extreme, Outlook and Resilience. Clicking on any of
the button will result in vertical lines displaying the
time in which these personality traits were displayed
toggle color circles to
show/hide emotions in the
zoom-in timeline
complete
timeline of
the user
zoom-in
view of the
timeline
over the
selected
time period
valence
and arousal
of the
tweets in
the time
period
31
tweet texts
in the
selected
time period
mostmentioned
keywords
in the
selected
time period
© 2015 IBM Corporation
Person Analysis — Personality, Trustingness, and Trustworthiness
Analyze the Big-5 personality trait for a Twitter profile
Analyze trustingness and trustworthiness for a Twitter profile
32
© 2015 IBM Corporation
Person Analysis — Setup
Method 1: enter screen name or
twitter id and click the search icon
Method 2: for Twitters that have been
analyzed before, select from the drop
down menu and click the search icon
33
© 2015 IBM Corporation
Person Analysis — Usage
hover over the circle on the time period
boundary to view the trend of each
personal characteristics which are color
coded as the upper-right corner
hover over a tweet(background color changed to orange) and the keywords
that are associated with valence and arousal in the tweet are highlighted in
bold font and shown on the left hand plot
34
© 2015 IBM Corporation
Target Discovery
Provide anomaly measurements of Twitter user profiles and visually assist analyst investigations
all users’ anomaly
measurements
(each node is a
user)
circular timeline view
of selected users
profiles of users
with highest
anomaly
measurements
timeline view of features extracted
from selected user profiles
35
© 2015 IBM Corporation
Target Discovery — Setup
Step 3: select the filtering range
for the selected metric by drag
and drop the min/max squares
on the value bar
Step 2: select a
metric for filtering
the user profiles
from the drop down
menu
Step 1: select a dataset
from the drop down menu
36
summary view
© 2015 IBM Corporation
Target Discovery — Usage (Global View)
reset the
view
zoom
in/out
select all users
in the view
each node is a Twitter user profile: size of the
node represents the number of followers of the
user, and the color of the node represents the
anomaly measurements of the user
click on the node to add the user to the
summary view
hover over a node to show the details of the
user
right-click on a node to annotate if
the user is anomalous (Yes) or not
(No) or clear annotation (Clear)
37
© 2015 IBM Corporation
Target Discovery — Usage (Summary View)
each circular
timeline represents
a user selected
from the global
view
the color of the centre represent
the anomaly score (if “Anomaly”
is selected) or the averaged
sentiment of all this user’s tweets
(if “Sentiment” is selected)
each perpendicular line to the circle represents
a) one tweet (if “Behavior” is selected): its length
represents the lifetime of the tweet and its color
represents its sentiment
b) z-score of one feature (if “Z-Score” is selected)
c) value of one feature(if “Feature” is selected)
38
© 2015 IBM Corporation
Target Discovery — Usage (Summary View)
when “Network” is selected, the
arrows between nodes
represent which users are
following which (outgoing node
is following the incoming node
39
© 2015 IBM Corporation
Anomaly Detection
Detect top anomalous re-tweet sequences
retweet sequence: color of each eclipse represents
the value of the selected feature of the retweeter
hover over each eclipse and the person
analysis result of the corresponding
retweeter will be shown on the right
40
© 2015 IBM Corporation
Anomaly Detection — Setup
click to select a dataset
41
© 2015 IBM Corporation