Influence of NaMo App on Twitter

IIIT Hyderabad
IIIT Hyderabad IIIT Hyderabad
Influence of NaMo App on Twitter
Shreya Sharma
M.Tech CSE
IIT Kanpur
shreyasa@iitk.ac.in
Supervised by:
Prof. Ponnurangam Kumaraguru
Prof. Amey Karkare
Why Social Media Analytics ?
● Approximately 51% of world’s population use social media
● On an average, user spend 2.5 hours daily
● Content generated by traditional media << Content generated by social media sites
● Used to spot trends and conversations and get information like sentiment, opinion, network formed
2
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
2019 Lok Sabha Elections
● Online election campaigns cost - 586 Millions
● 130 Million first time voters - 15% in 20s
● India’s first “WhatsApp Election” - 50,000 groups to spread campaign information
● First time :
○ Election Commission of India (ECI) issued social media guidelines
○ Candidates need to submit information about social media handles
● Most popular platform - Twitter
3
Campbell-Smith, Ualan and Bradshaw, Samantha, 2019, Global Cyber Troops Country Profile: India.
https://demtech.oii.ox.ac.uk/wp-content/uploads/sites/93/2019/05/India-Profile.pdf
1
1
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
NaMo App
● Centered around particular political party - BJP
● Extremely right leaning content
● Features
○ Activity points
○ Direct emails from prime minister
○ Built-in twitter-like network - My Network
○ Post, like, comment, and share
4
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Content shared from NaMo App
5
Shared from
MyNetwork -
annotated with
via MyNt tag
Shared from
feed - annotated
with via NaMo
App tag
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Motivation
After 2019 Indian Lok Sabha Elections, evidence were found demonstrating the growing capacity of cyber-troops, tasked with
manipulating public opinion online with geographically coordinated behaviours. Networks of paid workers and volunteers
disseminate sophisticated disinformation strategies across social media. NaMo App is created by one such IT firm which was
linked to fake facebook accounts.
● How does NaMo App affects the online conversation on more traditional networks such as Twitter?
● NaMo app consists of only biased content, so does it receives the same amount of engagement and the reaction on
Twitter?
6
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Campbell-Smith, Ualan and Bradshaw, Samantha, 2019, Global Cyber Troops Country Profile: India.
https://demtech.oii.ox.ac.uk/wp-content/uploads/sites/93/2019/05/India-Profile.pdf
1
1
NaMo App -> Twitter
Users Content
Reachability?
Type, Amount, Time?
Part of Echo chamber
or Neutral users ?
Users - Who, Type?
Which party ?
Influence
Geographical
location? cluster?
Similarity with NaMo
content?
How much? Quantify?
7
Can we characterise the users that post
NaMo content on Twitter?
● Part of Echo Chamber?
● Certain states form a larger part ?
● Difference between users that post
using NaMo app and who post
similar content on Twitter
How much influence does NaMo
content have on Twitter?
How much content on Twitter is
contributed by NaMo?
What type of content shared from NaMo
App to Twitter ?
Reachability of NaMo content on Twitter in
comparison to its similar content?
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Datasets
● Twitter Datasets:
○ 2019 Lok Sabha Election
■ 45 million tweets
■ Maintained by the PreCog group at IIIT Delhi / Hyderabad
○ Citizenship Amendment Act (CAA) protests
■ 1.2 million tweets
■ Collected by Neha Kumari, PhD at IIIT Delhi / Hyderabad
○ 2020 Delhi Election
■ 1.27 million users
■ Collected by the PreCog group at IIIT Delhi / Hyderabad
● NaMo App
○ Collected by Rohan Rajpal, B.Tech at IIIT Delhi / Hyderabad
● Used Twitter API for other Dataset
8
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Echo Chamber during 2019 Lok
Sabha Elections
What is an Echo Chamber ?
● Environments in which users’ opinions, or political leanings are reinforced due
to repeated interactions with peers that share similar tendencies and attitudes
● For our research, Echo Chamber is - the political leaning of the content of user
agrees with political leaning of the content of users from network
10
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
11
Network Analysis
Analyzed user’s followers,
following, and retweet
network
Content Analysis
Analyzed polarities of content
users produce and consume
Partitioning of Users
● To find political affiliations
● BJP vs Others
Filtering of Users
● Tweeted at least 50 tweets during
2019 Lok Sabha Election and 2020
Delhi Election
● Used most frequent hashtags
● 6037 common users
Methodology to identify Echo Chamber
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
12
NivaDuck Dataset
● Twitter handles of Indian
political leaders
● 243 users in common
Hashtag Usage
● Utilised hashtag usage as the
proxy for political leanings
● Annotated political hashtags
trended during 2019 Lok
Sabha Elections as pro-bjp,
anti-bjp, pro-congress,
anti-congress and neutral
User Metadata (Profile attributes)
● Twitter’s screen_name and their descriptions
● 474 users identified
Partition of
Users
Partitioning (identify political leaning) of Users
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
User Metadata - To partition the users
Screen_name:
● Keywords: bjp, modi, namo, congress, inc, cong
Description:
● Keywords: bjp, modi, namo, narendra modi, bhartiya janta
party, amit shah, amitshah, narendramodi, bjp4india,
rahul gandhi, congress, inc, priyanka gandhi, raga, shashi
tharoor
MODIfied_SKP BJP
Krish_BJP BJP
Shivam_INC INC
amitsoni_INC INC
Hardcore MODIJI fan. Former RSS
pracharak. District secretary BJP IT & Social
media Tirupattur DT. Tamil nadu.
BJP
Lawyer & Not associated with Congress in
any manner My tweets are my personal
views and still Rahul Gandhi is my leader
INC
13
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
*Manually checked all usernames and descriptions to ensure correct leaning
Hashtag usage
Calculated following ratios for number of hashtags for each
user:
● pro bjp ratio - A / E
● pro congress ratio - B / E
● anti bjp ratio - C / E
● anti congress ratio - D / E
● percent of annotated hashtags - ( E / F )
14
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
● number of annotated hashtags - E
● total hashtags (annotated and no-label) - F
● number of pro bjp hashtags - A
● number of pro congress hashtags - B
● number of anti bjp hashtags - C
● number of anti congress hashtags - D
After partitioning
● 4637 users politically annotated out of 6037 users
Users who have used utilised 0.1% of the annotated hashtags - allocated a leaning based on the ratio with the highest
value - assigned leaning to 3920 users
Content Analysis
Methodology
Assigned
● Hashtags belonging to pro-BJP - score 1
● Hashtags belonging to anti-BJP or pro-Congress - score 0
16
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Calculated the following values for each user :
● Production polarity : The average of the scores of all the hashtags
● Production variance : The variance in the scores of all the hashtags
● Consumption polarity : The average of the scores of all the hashtags that user is consuming from its network
● Consumption variance : The variance in the scores of all the hashtags that user is consuming from its network
Polarity vs Variance
17
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Little variation in polarity of content of users with an average polarity close to 0 or 1
High variation
for users with
polarity close
to 0.5
Higher polarity - pro-BJP
Lower polarity - anti-BJP/pro-congress
Consumption vs Production
● BJP users - Red
● Other users - Blue
● Most users are producing and
consuming content with similar
polarity
18
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Pro-BJP - Highly
dense
Pro-congress - spread
out
Network Analysis
Methodology
● 6037 users - 4637 users politically annotated
● Gephi - Network analysis tool
● Types of Network created - Follow, Following, Retweet
● Node Size - Degree/connections
● Node Color
○ BJP - Purple
○ Other (Anti-BJP, Congress) - Green
○ Unknown - Orange
20
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Network Analysis - Methodology
Step 1: Importing nodes and edges Step 2: Calculated graph modularity
Ref: Fast Unfolding of Communities in Large Networks
21
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Network Analysis - Methodology
Step 3:
Applied Force Atlas 2 layout for visualization
It performs:
● Scaling: Scales expansion of graph
● Dissuade hubs: Stronger repulsive forces to opposing hubs
● Prevent overlapping of nodes
22
Step 4: Color nodes according to political affiliation
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Follow Network
Community 1
Community 2
Community 3
% of users
Community # Pro-BJP Other Unknown
1 95.8% 0.3 3.9
2 8.8 53.4 37.8
3 9.8 12.8 77.4
23
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Influence of NaMo App on
Twitter
Terminologies/Definitions
● Seed users: The users who post content using NaMo App on Twitter
● Auxiliary users: The users who post tweets with content similar to that of tweets posted via NaMo App but does not use NaMo App for
posting them
● Affected users: Seed users + auxiliary users
● NaMo tweets: Tweets posted using NaMo App on Twitter
● Non-NaMo tweets: Tweets similar to NaMo tweets but not posted using NaMo App on Twitter
25
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Dataset Creation
● Collected tweets that have via MyNt or via NaMo App tag at the end
● Collected URLs of the images that are part of tweets that have via MyNt or via NaMo App tag
RQ 1.1 -
How much content on Twitter is contributed
by NaMo?
Methodology for Text Clustering
To get tweets similar to NaMo tweets:
● Removed words in the vocabulary with document frequency 1
● Removed tweets with length less than 5
● Performed K-Means clustering
● Only considered a match if euclidean distance of the normalized
vector from its cluster centroid was less than 0.45
27
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
28
Phashes
pHash Extraction using
ImageHash
Pairwise
comparisons
pHash-based Pairwise
Distance
Calculation
Clusters and images
of medoids
Clustering and medoid
calculation
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Methodology for Image Clustering
Example of a Cluster
29
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Results of Clustering - Matched Content
For Text
● 4170 Non-NaMo tweets matched to NaMo tweets
● 20500 Non-NaMo tweets matched to NaMo tweets (for CAA Dataset)
For Images
● 4705 unique Non-NaMo images matched to NaMo images
30
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Temporal Analysis
To get one-to-one mapping of NaMo tweets with its matched Non-NaMo tweets (tweets similar to NaMo
tweets):
● Calculated cosine distance of NaMo tweet with every Non-NaMo tweet
● Mapped a NaMo tweet to another Non-NaMo tweet with a maximum similarity score
● Compared the timestamps of NaMo tweets with Non-NaMo tweets
31
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Results of Temporal Analysis
After removing exact Non-NaMo tweets and considering only first occurrence,
32
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
First Occurence
(For 2019 Election Dataset, left with 940 tweets )
First Occurence
(For CAA Dataset, left with 2811 tweets)
RQ 1.2 -
What type of content is shared from
NaMo App to Twitter ?
Word Cloud (CAA Protests)
Tweets not posted using NaMo App
34
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Observed:
● Terms focusing on CAB protests such as ‘protests’,
‘indiaagainstcaa’, ‘anti’, ‘violence’, ‘ वरोध’, ‘muslim’,
‘protesting’, ‘students’, ‘police’, etc
Observed:
● Phrases in support of CAA bill such as
‘indiasupportscaa’, ‘caaclarified’, ‘isupportcaa’,
‘provisions’, etc
● Terms related to nationalism, diversity or places such
as ‘pakistan’, ‘northeast’, ‘amritsar’, ‘afghanistan, ‘sikhs’,
‘hindu’, ‘linguistic’, ‘bangladesh’, ‘kashmiri’, etc
35
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Tweets posted using NaMo App
Word Cloud (CAA Protests)
Word cloud for auxiliary users (2019 Lok Sabha Election)
Tweets similar to tweets posted using NaMo App Tweets not similar to posts posted using NaMo App
36
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Word cloud for seed users (2019 Lok Sabha Election)
Tweets posted using NaMo App Tweets not posted using NaMo App
37
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Seed users:
Posted using NaMo App
● Pro-BJP phrases such as ‘deshkagauravmodi’,
‘indiavotesfornamo’, ‘aayegatomodihi’
● Terms related to development or schemes such as ‘benefits’,
‘employment’, ‘aadhar’, ‘transparent’, ‘opportunities’,
‘betterbharat’, ‘empower’, ‘women’, ‘middleman’, etc
Not posted using NaMo App
● Mostly posting tweets containing terms related to criticising
other parties such as ‘gandhi’, ‘congress’, ‘ममता’, ‘पप्पू’, ‘क
े जरीवाल’,
‘दीदी’, ‘ वपक्ष’
Auxiliary users:
● Only posting about pro-BJP phrases such as
‘deshkagauravmodi’, ‘modiaanewalahai’,
‘indiawantsmodiagain’
Observations (2019 Lok Sabha Elections)
38
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Conclusions of RQ 1
● Through temporal analysis, concluded
○ Tweets made using the NaMo App were already present on Twitter
○ No new information was coming from NaMo App
○ Users who uses NaMo App posted content on Twitter first
● Through content analysis, concluded that
○ Affected users may be highly pro-BJP
■ Auxiliary users - only praising BJP
■ Seed users - apart from pro-bjp phrases, criticising other political parties
○ Tweets
■ posted from NaMo App - about policies, clarifying the govt, BJP’s benefits
39
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
RQ 2 -
How much influence does NaMo content
have on Twitter?
Hawkes Process
To show NaMo App is the cause of some posting activity on Twitter
● Event on one process can cause a response on other processes, increasing the probability of an event occurring on other
processes
● Can be used for different types of content posted on web communities
● Each web community has its own rates for posting the images as well as some influence due to other social media sites
● In our setting,
○ events -> posting of an image
○ processes -> web communities
41
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Results of Hawkes Process
∆t = 24 hrs NaMo Twitter
NaMo 0.25667107 0.1334780
3
Twitter 0.18027061 0.2567351
6
∆t = 48 hrs NaMo Twitter
NaMo 0.25685813 0.1402505
Twitter 0.54783613 0.2552396
4
Weight Matrix - Influence matrix
● Amount of interaction from one process to another
● Weight value act as a rate parameter in the Poisson process
● Can be interpreted as the expected number of events caused on
a web community due to another web community
● For our research,
○ expected events - posting of an image from one platform
to another
∆t - Indicates that an event on one platform can cause an event on other
platform within the given time window
42
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Conclusions of RQ 2
● Observed similar values for all the time periods
● Twitter to Twitter - Retweeting activity on twitter can be attributed to the same image being shared on the same platform
● NaMo App to NaMo App - Value can be used to indicate that the same image posted by other user
● Influence:
○ Twitter to NaMo App > NaMo App to Twitter
● Content is first shared on Twitter, then circulated or posted to NaMo App through other means
● Similar results were obtained through temporal analysis
43
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
RQ 3 -
Can we characterise the users that post
NaMo content on Twitter?
RQ 3.1 - Are users part of an Echo Chamber on
Twitter?
● Hashtags belonging to pro-BJP - score 1
● Hashtags belonging to anti-BJP/pro-Congress - score 0
● Properties considered:
a. Production polarity
b. Production variance
c. Consumption polarity
d. Consumption variance
45
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Polarity vs Variance
46
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Consumption vs Production
47
Consumption and Production
polarity close to 1 - Pro-BJP users
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
RQ 3.2 - Do users of certain states form a larger
part of the user group than general users?
● Obtained locations of all seed and auxiliary users from their accounts
● Mapped city to state
● Calculated count of users for each state
48
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Fraction of Users (2019 Lok Sabha Election)
State Fraction of Affected Users Fraction of General Users
Uttar Pradesh 0.171811 0.140201
NCT of Delhi 0.156173 0.184110
Gujarat 0.128189 0.057137
Maharashtra 0.109671 0.149511
Rajasthan 0.062963 0.057609
49
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Affected user count 🔺
Affected user count🔻
RQ 3.3 - Is there any difference between seed
and auxiliary users?
● Created Word Cloud and calculated Odd-Ratios of twitter descriptions of Seed and Auxiliary users
● Higher Odd Ratio for a bigram or trigram for a given class indicates that the bigram or trigram has a close
relationship with that class
50
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Word Cloud of twitter descriptions (twitter bio)
Seed Users Auxiliary Users
51
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Odd Ratios of Bigrams (2019 Lok Sabha Elections)
Seed Users Auxiliary Users
52
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Conclusions of RQ 3
● RQ3.1 - Affected users generate and consume tweets reflecting their pro-BJP ideology - part of an echo chamber
● RQ3.2 - Affected user group are more likely to belong to a heavily pro-BJP state
● RQ3.3 - Possibility of auxiliary users to hold relations with members of BJP party , whereas seed users are likely to
be supporters or lower-level worker of BJP and associated groups
53
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
RQ 4 -
What is the reachability of NaMo content on
Twitter in comparison to non-NaMo content?
Correlation graphs normalized by followers count
55
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
CDF Graphs
56
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Conclusions of RQ 4
● Content posted through the NaMo App may not have much reachability. Thus, does not
affecting the larger discourse on Twitter.
● Observed similar results for CAA dataset
57
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Conclusions
● Less influence of NaMo App on Twitter - not affecting the larger discourse and shows its inability to change the narrative on Twitter
● Association of auxiliary users with members of BJP party.
● Seed users (likely to be BJP workers) are responsible for cross posting content, were seen praising BJP party and demeaning other
political parties.
● Affected users are part of online echo chambers - unable to break into a wider and more diverse audience
● Limited reachability of the NaMo content on Twitter
58
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
Future Work
● To identify users that BJP should target to increase its reach or influence through NaMo App
● A deeper content analysis to identify topics which are often cross posted
● If exists in future, one can incorporate party apps and its data to compare their visibility
59
CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
60
Limitations and Challenges:
● Naive Political Classification
○ Considered only hashtags, and not the context of hashtags
■ Sarcasm tweets/Parity users - intent of tweet not aligning with intent of involved hashtags - may get allocated a
wrong political leaning
○ Classification was done only for binary (BJP vs Others) on the basis of top 600 annotated hashtags - while this could have
also been done in multi class problem (Pro-BJP, Anti-BJP, Pro-Cong, Anti-Cong or other political party)
● Users with more neutral hashtags and less political hashtags - will get assigned a leaning on the basis of political hashtags only
● To get a better understanding of count of affected users- we could also compare these values by considering the tweets not from lok
sabha elections posted by users from these states
● Code-Mixed Data
Thank You
1 of 61

Recommended

Responsible & Safe AI: #LegalBias #Inconsistency #BiasinLLMs #MultiModalBias by
Responsible & Safe AI: #LegalBias #Inconsistency #BiasinLLMs #MultiModalBiasResponsible & Safe AI: #LegalBias #Inconsistency #BiasinLLMs #MultiModalBias
Responsible & Safe AI: #LegalBias #Inconsistency #BiasinLLMs #MultiModalBiasIIIT Hyderabad
101 views56 slides
Identify, Inspect and Intervene Multimodal Fake News by
Identify, Inspect and Intervene Multimodal Fake NewsIdentify, Inspect and Intervene Multimodal Fake News
Identify, Inspect and Intervene Multimodal Fake NewsIIIT Hyderabad
117 views120 slides
#ChatGPT #ResponsibleAI by
#ChatGPT #ResponsibleAI#ChatGPT #ResponsibleAI
#ChatGPT #ResponsibleAIIIIT Hyderabad
331 views21 slides
Data Science for Social Good: #MentalHealth #CodeMix #LegalNLP #AISafety by
Data Science for Social Good: #MentalHealth #CodeMix #LegalNLP #AISafetyData Science for Social Good: #MentalHealth #CodeMix #LegalNLP #AISafety
Data Science for Social Good: #MentalHealth #CodeMix #LegalNLP #AISafetyIIIT Hyderabad
46 views38 slides
It is our choices, Harry, that show what we truly are, far more than our abil... by
It is our choices, Harry, that show what we truly are, far more than our abil...It is our choices, Harry, that show what we truly are, far more than our abil...
It is our choices, Harry, that show what we truly are, far more than our abil...IIIT Hyderabad
86 views56 slides
Beyond the Surface: A Computational Exploration of Linguistic Ambiguity by
Beyond the Surface: A Computational Exploration of Linguistic AmbiguityBeyond the Surface: A Computational Exploration of Linguistic Ambiguity
Beyond the Surface: A Computational Exploration of Linguistic AmbiguityIIIT Hyderabad
57 views79 slides

More Related Content

More from IIIT Hyderabad

Data Science for Social Good: #LegalNLP #AlgorithmicBias by
Data Science for Social Good: #LegalNLP #AlgorithmicBiasData Science for Social Good: #LegalNLP #AlgorithmicBias
Data Science for Social Good: #LegalNLP #AlgorithmicBiasIIIT Hyderabad
52 views37 slides
Social Computing Research in India by
Social Computing Research in IndiaSocial Computing Research in India
Social Computing Research in IndiaIIIT Hyderabad
84 views46 slides
Social Computing Research in India by
Social Computing Research in IndiaSocial Computing Research in India
Social Computing Research in IndiaIIIT Hyderabad
235 views53 slides
Modeling Online User Interactions and their Offline effects on Socio-Technica... by
Modeling Online User Interactions and their Offline effects on Socio-Technica...Modeling Online User Interactions and their Offline effects on Socio-Technica...
Modeling Online User Interactions and their Offline effects on Socio-Technica...IIIT Hyderabad
119 views305 slides
Privacy. Winter School on “Topics in Digital Trust”. IIT Bombay by
Privacy. Winter School on “Topics in Digital Trust”. IIT BombayPrivacy. Winter School on “Topics in Digital Trust”. IIT Bombay
Privacy. Winter School on “Topics in Digital Trust”. IIT BombayIIIT Hyderabad
54 views107 slides
It is our choices, Harry, that show what we truly are, far more than our abil... by
It is our choices, Harry, that show what we truly are, far more than our abil...It is our choices, Harry, that show what we truly are, far more than our abil...
It is our choices, Harry, that show what we truly are, far more than our abil...IIIT Hyderabad
56 views41 slides

More from IIIT Hyderabad (20)

Data Science for Social Good: #LegalNLP #AlgorithmicBias by IIIT Hyderabad
Data Science for Social Good: #LegalNLP #AlgorithmicBiasData Science for Social Good: #LegalNLP #AlgorithmicBias
Data Science for Social Good: #LegalNLP #AlgorithmicBias
IIIT Hyderabad 52 views
Social Computing Research in India by IIIT Hyderabad
Social Computing Research in IndiaSocial Computing Research in India
Social Computing Research in India
IIIT Hyderabad 235 views
Modeling Online User Interactions and their Offline effects on Socio-Technica... by IIIT Hyderabad
Modeling Online User Interactions and their Offline effects on Socio-Technica...Modeling Online User Interactions and their Offline effects on Socio-Technica...
Modeling Online User Interactions and their Offline effects on Socio-Technica...
IIIT Hyderabad 119 views
Privacy. Winter School on “Topics in Digital Trust”. IIT Bombay by IIIT Hyderabad
Privacy. Winter School on “Topics in Digital Trust”. IIT BombayPrivacy. Winter School on “Topics in Digital Trust”. IIT Bombay
Privacy. Winter School on “Topics in Digital Trust”. IIT Bombay
IIIT Hyderabad 54 views
It is our choices, Harry, that show what we truly are, far more than our abil... by IIIT Hyderabad
It is our choices, Harry, that show what we truly are, far more than our abil...It is our choices, Harry, that show what we truly are, far more than our abil...
It is our choices, Harry, that show what we truly are, far more than our abil...
IIIT Hyderabad 56 views
It is our choices, Harry, that show what we truly are, far more than our abil... by IIIT Hyderabad
It is our choices, Harry, that show what we truly are, far more than our abil...It is our choices, Harry, that show what we truly are, far more than our abil...
It is our choices, Harry, that show what we truly are, far more than our abil...
IIIT Hyderabad 389 views
Leveraging Social Media for Financial Advice by IIIT Hyderabad
Leveraging Social Media for Financial AdviceLeveraging Social Media for Financial Advice
Leveraging Social Media for Financial Advice
IIIT Hyderabad 96 views
Development of Stress Induction and Detection System to Study its Effect on B... by IIIT Hyderabad
Development of Stress Induction and Detection System to Study its Effect on B...Development of Stress Induction and Detection System to Study its Effect on B...
Development of Stress Induction and Detection System to Study its Effect on B...
IIIT Hyderabad 107 views
A Framework for Automatic Question Answering in Indian Languages by IIIT Hyderabad
A Framework for Automatic Question Answering in Indian LanguagesA Framework for Automatic Question Answering in Indian Languages
A Framework for Automatic Question Answering in Indian Languages
IIIT Hyderabad 191 views
A Framework For Automatic Question Answering in Indian Languages by IIIT Hyderabad
A Framework For Automatic Question Answering in Indian LanguagesA Framework For Automatic Question Answering in Indian Languages
A Framework For Automatic Question Answering in Indian Languages
IIIT Hyderabad 26 views
Exposing, Examining and Intervening Fake News by IIIT Hyderabad
Exposing, Examining and Intervening Fake NewsExposing, Examining and Intervening Fake News
Exposing, Examining and Intervening Fake News
IIIT Hyderabad 98 views
It's MY JOB: Identifying and Improving Content Quality for Online recruitmen... by IIIT Hyderabad
 It's MY JOB: Identifying and Improving Content Quality for Online recruitmen... It's MY JOB: Identifying and Improving Content Quality for Online recruitmen...
It's MY JOB: Identifying and Improving Content Quality for Online recruitmen...
IIIT Hyderabad 52 views
De-anonymizing, Preserving and Democratizing Data Privacy and Ownership by IIIT Hyderabad
De-anonymizing, Preserving and Democratizing Data Privacy and OwnershipDe-anonymizing, Preserving and Democratizing Data Privacy and Ownership
De-anonymizing, Preserving and Democratizing Data Privacy and Ownership
IIIT Hyderabad 53 views
Justice Delayed is Justice Denied: Enabling Legal Artificial Intelligence via... by IIIT Hyderabad
Justice Delayed is Justice Denied: Enabling Legal Artificial Intelligence via...Justice Delayed is Justice Denied: Enabling Legal Artificial Intelligence via...
Justice Delayed is Justice Denied: Enabling Legal Artificial Intelligence via...
IIIT Hyderabad 139 views
NLP / Language Research at Precog by IIIT Hyderabad
NLP / Language Research at PrecogNLP / Language Research at Precog
NLP / Language Research at Precog
IIIT Hyderabad 217 views
“It is our choices, Harry, that show what we truly are, far more than our abi... by IIIT Hyderabad
“It is our choices, Harry, that show what we truly are, far more than our abi...“It is our choices, Harry, that show what we truly are, far more than our abi...
“It is our choices, Harry, that show what we truly are, far more than our abi...
IIIT Hyderabad 227 views
What's Kooking? Characterizing India's Emerging Social Network, Koo by IIIT Hyderabad
What's Kooking? Characterizing India's Emerging Social Network, KooWhat's Kooking? Characterizing India's Emerging Social Network, Koo
What's Kooking? Characterizing India's Emerging Social Network, Koo
IIIT Hyderabad 527 views
Code Mixing computationally bahut challenging hai by IIIT Hyderabad
Code Mixing computationally bahut challenging haiCode Mixing computationally bahut challenging hai
Code Mixing computationally bahut challenging hai
IIIT Hyderabad 353 views
Roadmap for Initiating Joint Collaborations by IIIT Hyderabad
Roadmap for Initiating Joint CollaborationsRoadmap for Initiating Joint Collaborations
Roadmap for Initiating Joint Collaborations
IIIT Hyderabad 295 views

Recently uploaded

Art of Writing Research article slide share.pptx by
Art of Writing Research article slide share.pptxArt of Writing Research article slide share.pptx
Art of Writing Research article slide share.pptxsureshc91
14 views42 slides
Object Oriented Programming with JAVA by
Object Oriented Programming with JAVAObject Oriented Programming with JAVA
Object Oriented Programming with JAVADemian Antony D'Mello
64 views28 slides
String.pptx by
String.pptxString.pptx
String.pptxAnanthi Palanisamy
47 views24 slides
DevOps to DevSecOps: Enhancing Software Security Throughout The Development L... by
DevOps to DevSecOps: Enhancing Software Security Throughout The Development L...DevOps to DevSecOps: Enhancing Software Security Throughout The Development L...
DevOps to DevSecOps: Enhancing Software Security Throughout The Development L...Anowar Hossain
10 views34 slides
802.11 Computer Networks by
802.11 Computer Networks802.11 Computer Networks
802.11 Computer NetworksTusharChoudhary72015
9 views33 slides
performance uploading.pptx by
performance uploading.pptxperformance uploading.pptx
performance uploading.pptxSanthiS10
7 views18 slides

Recently uploaded(20)

Art of Writing Research article slide share.pptx by sureshc91
Art of Writing Research article slide share.pptxArt of Writing Research article slide share.pptx
Art of Writing Research article slide share.pptx
sureshc9114 views
DevOps to DevSecOps: Enhancing Software Security Throughout The Development L... by Anowar Hossain
DevOps to DevSecOps: Enhancing Software Security Throughout The Development L...DevOps to DevSecOps: Enhancing Software Security Throughout The Development L...
DevOps to DevSecOps: Enhancing Software Security Throughout The Development L...
Anowar Hossain10 views
performance uploading.pptx by SanthiS10
performance uploading.pptxperformance uploading.pptx
performance uploading.pptx
SanthiS107 views
STUDY OF SMART MATERIALS USED IN CONSTRUCTION-1.pptx by AnnieRachelJohn
STUDY OF SMART MATERIALS USED IN CONSTRUCTION-1.pptxSTUDY OF SMART MATERIALS USED IN CONSTRUCTION-1.pptx
STUDY OF SMART MATERIALS USED IN CONSTRUCTION-1.pptx
AnnieRachelJohn25 views
A multi-microcontroller-based hardware for deploying Tiny machine learning mo... by IJECEIAES
A multi-microcontroller-based hardware for deploying Tiny machine learning mo...A multi-microcontroller-based hardware for deploying Tiny machine learning mo...
A multi-microcontroller-based hardware for deploying Tiny machine learning mo...
IJECEIAES10 views
Machine Element II Course outline.pdf by odatadese1
Machine Element II Course outline.pdfMachine Element II Course outline.pdf
Machine Element II Course outline.pdf
odatadese16 views
13_DVD_Latch-up_prevention.pdf by Usha Mehta
13_DVD_Latch-up_prevention.pdf13_DVD_Latch-up_prevention.pdf
13_DVD_Latch-up_prevention.pdf
Usha Mehta9 views
Design and analysis of a new undergraduate Computer Engineering degree – a me... by WaelBadawy6
Design and analysis of a new undergraduate Computer Engineering degree – a me...Design and analysis of a new undergraduate Computer Engineering degree – a me...
Design and analysis of a new undergraduate Computer Engineering degree – a me...
WaelBadawy652 views
Performance of Back-to-Back Mechanically Stabilized Earth Walls Supporting th... by ahmedmesaiaoun
Performance of Back-to-Back Mechanically Stabilized Earth Walls Supporting th...Performance of Back-to-Back Mechanically Stabilized Earth Walls Supporting th...
Performance of Back-to-Back Mechanically Stabilized Earth Walls Supporting th...
ahmedmesaiaoun12 views
7_DVD_Combinational_MOS_Logic_Circuits.pdf by Usha Mehta
7_DVD_Combinational_MOS_Logic_Circuits.pdf7_DVD_Combinational_MOS_Logic_Circuits.pdf
7_DVD_Combinational_MOS_Logic_Circuits.pdf
Usha Mehta50 views
Update 42 models(Diode/General ) in SPICE PARK(DEC2023) by Tsuyoshi Horigome
Update 42 models(Diode/General ) in SPICE PARK(DEC2023)Update 42 models(Diode/General ) in SPICE PARK(DEC2023)
Update 42 models(Diode/General ) in SPICE PARK(DEC2023)
Extensions of Time - Contract Management by brainquisitive
Extensions of Time - Contract ManagementExtensions of Time - Contract Management
Extensions of Time - Contract Management
brainquisitive15 views

Influence of NaMo App on Twitter

  • 1. Influence of NaMo App on Twitter Shreya Sharma M.Tech CSE IIT Kanpur shreyasa@iitk.ac.in Supervised by: Prof. Ponnurangam Kumaraguru Prof. Amey Karkare
  • 2. Why Social Media Analytics ? ● Approximately 51% of world’s population use social media ● On an average, user spend 2.5 hours daily ● Content generated by traditional media << Content generated by social media sites ● Used to spot trends and conversations and get information like sentiment, opinion, network formed 2 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 3. 2019 Lok Sabha Elections ● Online election campaigns cost - 586 Millions ● 130 Million first time voters - 15% in 20s ● India’s first “WhatsApp Election” - 50,000 groups to spread campaign information ● First time : ○ Election Commission of India (ECI) issued social media guidelines ○ Candidates need to submit information about social media handles ● Most popular platform - Twitter 3 Campbell-Smith, Ualan and Bradshaw, Samantha, 2019, Global Cyber Troops Country Profile: India. https://demtech.oii.ox.ac.uk/wp-content/uploads/sites/93/2019/05/India-Profile.pdf 1 1 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 4. NaMo App ● Centered around particular political party - BJP ● Extremely right leaning content ● Features ○ Activity points ○ Direct emails from prime minister ○ Built-in twitter-like network - My Network ○ Post, like, comment, and share 4 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 5. Content shared from NaMo App 5 Shared from MyNetwork - annotated with via MyNt tag Shared from feed - annotated with via NaMo App tag CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 6. Motivation After 2019 Indian Lok Sabha Elections, evidence were found demonstrating the growing capacity of cyber-troops, tasked with manipulating public opinion online with geographically coordinated behaviours. Networks of paid workers and volunteers disseminate sophisticated disinformation strategies across social media. NaMo App is created by one such IT firm which was linked to fake facebook accounts. ● How does NaMo App affects the online conversation on more traditional networks such as Twitter? ● NaMo app consists of only biased content, so does it receives the same amount of engagement and the reaction on Twitter? 6 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation Campbell-Smith, Ualan and Bradshaw, Samantha, 2019, Global Cyber Troops Country Profile: India. https://demtech.oii.ox.ac.uk/wp-content/uploads/sites/93/2019/05/India-Profile.pdf 1 1
  • 7. NaMo App -> Twitter Users Content Reachability? Type, Amount, Time? Part of Echo chamber or Neutral users ? Users - Who, Type? Which party ? Influence Geographical location? cluster? Similarity with NaMo content? How much? Quantify? 7 Can we characterise the users that post NaMo content on Twitter? ● Part of Echo Chamber? ● Certain states form a larger part ? ● Difference between users that post using NaMo app and who post similar content on Twitter How much influence does NaMo content have on Twitter? How much content on Twitter is contributed by NaMo? What type of content shared from NaMo App to Twitter ? Reachability of NaMo content on Twitter in comparison to its similar content? CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 8. Datasets ● Twitter Datasets: ○ 2019 Lok Sabha Election ■ 45 million tweets ■ Maintained by the PreCog group at IIIT Delhi / Hyderabad ○ Citizenship Amendment Act (CAA) protests ■ 1.2 million tweets ■ Collected by Neha Kumari, PhD at IIIT Delhi / Hyderabad ○ 2020 Delhi Election ■ 1.27 million users ■ Collected by the PreCog group at IIIT Delhi / Hyderabad ● NaMo App ○ Collected by Rohan Rajpal, B.Tech at IIIT Delhi / Hyderabad ● Used Twitter API for other Dataset 8 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 9. Echo Chamber during 2019 Lok Sabha Elections
  • 10. What is an Echo Chamber ? ● Environments in which users’ opinions, or political leanings are reinforced due to repeated interactions with peers that share similar tendencies and attitudes ● For our research, Echo Chamber is - the political leaning of the content of user agrees with political leaning of the content of users from network 10 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 11. 11 Network Analysis Analyzed user’s followers, following, and retweet network Content Analysis Analyzed polarities of content users produce and consume Partitioning of Users ● To find political affiliations ● BJP vs Others Filtering of Users ● Tweeted at least 50 tweets during 2019 Lok Sabha Election and 2020 Delhi Election ● Used most frequent hashtags ● 6037 common users Methodology to identify Echo Chamber CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 12. 12 NivaDuck Dataset ● Twitter handles of Indian political leaders ● 243 users in common Hashtag Usage ● Utilised hashtag usage as the proxy for political leanings ● Annotated political hashtags trended during 2019 Lok Sabha Elections as pro-bjp, anti-bjp, pro-congress, anti-congress and neutral User Metadata (Profile attributes) ● Twitter’s screen_name and their descriptions ● 474 users identified Partition of Users Partitioning (identify political leaning) of Users CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 13. User Metadata - To partition the users Screen_name: ● Keywords: bjp, modi, namo, congress, inc, cong Description: ● Keywords: bjp, modi, namo, narendra modi, bhartiya janta party, amit shah, amitshah, narendramodi, bjp4india, rahul gandhi, congress, inc, priyanka gandhi, raga, shashi tharoor MODIfied_SKP BJP Krish_BJP BJP Shivam_INC INC amitsoni_INC INC Hardcore MODIJI fan. Former RSS pracharak. District secretary BJP IT & Social media Tirupattur DT. Tamil nadu. BJP Lawyer & Not associated with Congress in any manner My tweets are my personal views and still Rahul Gandhi is my leader INC 13 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation *Manually checked all usernames and descriptions to ensure correct leaning
  • 14. Hashtag usage Calculated following ratios for number of hashtags for each user: ● pro bjp ratio - A / E ● pro congress ratio - B / E ● anti bjp ratio - C / E ● anti congress ratio - D / E ● percent of annotated hashtags - ( E / F ) 14 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation ● number of annotated hashtags - E ● total hashtags (annotated and no-label) - F ● number of pro bjp hashtags - A ● number of pro congress hashtags - B ● number of anti bjp hashtags - C ● number of anti congress hashtags - D After partitioning ● 4637 users politically annotated out of 6037 users Users who have used utilised 0.1% of the annotated hashtags - allocated a leaning based on the ratio with the highest value - assigned leaning to 3920 users
  • 16. Methodology Assigned ● Hashtags belonging to pro-BJP - score 1 ● Hashtags belonging to anti-BJP or pro-Congress - score 0 16 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation Calculated the following values for each user : ● Production polarity : The average of the scores of all the hashtags ● Production variance : The variance in the scores of all the hashtags ● Consumption polarity : The average of the scores of all the hashtags that user is consuming from its network ● Consumption variance : The variance in the scores of all the hashtags that user is consuming from its network
  • 17. Polarity vs Variance 17 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation Little variation in polarity of content of users with an average polarity close to 0 or 1 High variation for users with polarity close to 0.5 Higher polarity - pro-BJP Lower polarity - anti-BJP/pro-congress
  • 18. Consumption vs Production ● BJP users - Red ● Other users - Blue ● Most users are producing and consuming content with similar polarity 18 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation Pro-BJP - Highly dense Pro-congress - spread out
  • 20. Methodology ● 6037 users - 4637 users politically annotated ● Gephi - Network analysis tool ● Types of Network created - Follow, Following, Retweet ● Node Size - Degree/connections ● Node Color ○ BJP - Purple ○ Other (Anti-BJP, Congress) - Green ○ Unknown - Orange 20 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 21. Network Analysis - Methodology Step 1: Importing nodes and edges Step 2: Calculated graph modularity Ref: Fast Unfolding of Communities in Large Networks 21 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 22. Network Analysis - Methodology Step 3: Applied Force Atlas 2 layout for visualization It performs: ● Scaling: Scales expansion of graph ● Dissuade hubs: Stronger repulsive forces to opposing hubs ● Prevent overlapping of nodes 22 Step 4: Color nodes according to political affiliation CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 23. Follow Network Community 1 Community 2 Community 3 % of users Community # Pro-BJP Other Unknown 1 95.8% 0.3 3.9 2 8.8 53.4 37.8 3 9.8 12.8 77.4 23 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 24. Influence of NaMo App on Twitter
  • 25. Terminologies/Definitions ● Seed users: The users who post content using NaMo App on Twitter ● Auxiliary users: The users who post tweets with content similar to that of tweets posted via NaMo App but does not use NaMo App for posting them ● Affected users: Seed users + auxiliary users ● NaMo tweets: Tweets posted using NaMo App on Twitter ● Non-NaMo tweets: Tweets similar to NaMo tweets but not posted using NaMo App on Twitter 25 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation Dataset Creation ● Collected tweets that have via MyNt or via NaMo App tag at the end ● Collected URLs of the images that are part of tweets that have via MyNt or via NaMo App tag
  • 26. RQ 1.1 - How much content on Twitter is contributed by NaMo?
  • 27. Methodology for Text Clustering To get tweets similar to NaMo tweets: ● Removed words in the vocabulary with document frequency 1 ● Removed tweets with length less than 5 ● Performed K-Means clustering ● Only considered a match if euclidean distance of the normalized vector from its cluster centroid was less than 0.45 27 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 28. 28 Phashes pHash Extraction using ImageHash Pairwise comparisons pHash-based Pairwise Distance Calculation Clusters and images of medoids Clustering and medoid calculation CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation Methodology for Image Clustering
  • 29. Example of a Cluster 29 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 30. Results of Clustering - Matched Content For Text ● 4170 Non-NaMo tweets matched to NaMo tweets ● 20500 Non-NaMo tweets matched to NaMo tweets (for CAA Dataset) For Images ● 4705 unique Non-NaMo images matched to NaMo images 30 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 31. Temporal Analysis To get one-to-one mapping of NaMo tweets with its matched Non-NaMo tweets (tweets similar to NaMo tweets): ● Calculated cosine distance of NaMo tweet with every Non-NaMo tweet ● Mapped a NaMo tweet to another Non-NaMo tweet with a maximum similarity score ● Compared the timestamps of NaMo tweets with Non-NaMo tweets 31 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 32. Results of Temporal Analysis After removing exact Non-NaMo tweets and considering only first occurrence, 32 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation First Occurence (For 2019 Election Dataset, left with 940 tweets ) First Occurence (For CAA Dataset, left with 2811 tweets)
  • 33. RQ 1.2 - What type of content is shared from NaMo App to Twitter ?
  • 34. Word Cloud (CAA Protests) Tweets not posted using NaMo App 34 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation Observed: ● Terms focusing on CAB protests such as ‘protests’, ‘indiaagainstcaa’, ‘anti’, ‘violence’, ‘ वरोध’, ‘muslim’, ‘protesting’, ‘students’, ‘police’, etc
  • 35. Observed: ● Phrases in support of CAA bill such as ‘indiasupportscaa’, ‘caaclarified’, ‘isupportcaa’, ‘provisions’, etc ● Terms related to nationalism, diversity or places such as ‘pakistan’, ‘northeast’, ‘amritsar’, ‘afghanistan, ‘sikhs’, ‘hindu’, ‘linguistic’, ‘bangladesh’, ‘kashmiri’, etc 35 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation Tweets posted using NaMo App Word Cloud (CAA Protests)
  • 36. Word cloud for auxiliary users (2019 Lok Sabha Election) Tweets similar to tweets posted using NaMo App Tweets not similar to posts posted using NaMo App 36 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 37. Word cloud for seed users (2019 Lok Sabha Election) Tweets posted using NaMo App Tweets not posted using NaMo App 37 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 38. Seed users: Posted using NaMo App ● Pro-BJP phrases such as ‘deshkagauravmodi’, ‘indiavotesfornamo’, ‘aayegatomodihi’ ● Terms related to development or schemes such as ‘benefits’, ‘employment’, ‘aadhar’, ‘transparent’, ‘opportunities’, ‘betterbharat’, ‘empower’, ‘women’, ‘middleman’, etc Not posted using NaMo App ● Mostly posting tweets containing terms related to criticising other parties such as ‘gandhi’, ‘congress’, ‘ममता’, ‘पप्पू’, ‘क े जरीवाल’, ‘दीदी’, ‘ वपक्ष’ Auxiliary users: ● Only posting about pro-BJP phrases such as ‘deshkagauravmodi’, ‘modiaanewalahai’, ‘indiawantsmodiagain’ Observations (2019 Lok Sabha Elections) 38 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 39. Conclusions of RQ 1 ● Through temporal analysis, concluded ○ Tweets made using the NaMo App were already present on Twitter ○ No new information was coming from NaMo App ○ Users who uses NaMo App posted content on Twitter first ● Through content analysis, concluded that ○ Affected users may be highly pro-BJP ■ Auxiliary users - only praising BJP ■ Seed users - apart from pro-bjp phrases, criticising other political parties ○ Tweets ■ posted from NaMo App - about policies, clarifying the govt, BJP’s benefits 39 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 40. RQ 2 - How much influence does NaMo content have on Twitter?
  • 41. Hawkes Process To show NaMo App is the cause of some posting activity on Twitter ● Event on one process can cause a response on other processes, increasing the probability of an event occurring on other processes ● Can be used for different types of content posted on web communities ● Each web community has its own rates for posting the images as well as some influence due to other social media sites ● In our setting, ○ events -> posting of an image ○ processes -> web communities 41 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 42. Results of Hawkes Process ∆t = 24 hrs NaMo Twitter NaMo 0.25667107 0.1334780 3 Twitter 0.18027061 0.2567351 6 ∆t = 48 hrs NaMo Twitter NaMo 0.25685813 0.1402505 Twitter 0.54783613 0.2552396 4 Weight Matrix - Influence matrix ● Amount of interaction from one process to another ● Weight value act as a rate parameter in the Poisson process ● Can be interpreted as the expected number of events caused on a web community due to another web community ● For our research, ○ expected events - posting of an image from one platform to another ∆t - Indicates that an event on one platform can cause an event on other platform within the given time window 42 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 43. Conclusions of RQ 2 ● Observed similar values for all the time periods ● Twitter to Twitter - Retweeting activity on twitter can be attributed to the same image being shared on the same platform ● NaMo App to NaMo App - Value can be used to indicate that the same image posted by other user ● Influence: ○ Twitter to NaMo App > NaMo App to Twitter ● Content is first shared on Twitter, then circulated or posted to NaMo App through other means ● Similar results were obtained through temporal analysis 43 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 44. RQ 3 - Can we characterise the users that post NaMo content on Twitter?
  • 45. RQ 3.1 - Are users part of an Echo Chamber on Twitter? ● Hashtags belonging to pro-BJP - score 1 ● Hashtags belonging to anti-BJP/pro-Congress - score 0 ● Properties considered: a. Production polarity b. Production variance c. Consumption polarity d. Consumption variance 45 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 46. Polarity vs Variance 46 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 47. Consumption vs Production 47 Consumption and Production polarity close to 1 - Pro-BJP users CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 48. RQ 3.2 - Do users of certain states form a larger part of the user group than general users? ● Obtained locations of all seed and auxiliary users from their accounts ● Mapped city to state ● Calculated count of users for each state 48 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 49. Fraction of Users (2019 Lok Sabha Election) State Fraction of Affected Users Fraction of General Users Uttar Pradesh 0.171811 0.140201 NCT of Delhi 0.156173 0.184110 Gujarat 0.128189 0.057137 Maharashtra 0.109671 0.149511 Rajasthan 0.062963 0.057609 49 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation Affected user count 🔺 Affected user count🔻
  • 50. RQ 3.3 - Is there any difference between seed and auxiliary users? ● Created Word Cloud and calculated Odd-Ratios of twitter descriptions of Seed and Auxiliary users ● Higher Odd Ratio for a bigram or trigram for a given class indicates that the bigram or trigram has a close relationship with that class 50 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 51. Word Cloud of twitter descriptions (twitter bio) Seed Users Auxiliary Users 51 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 52. Odd Ratios of Bigrams (2019 Lok Sabha Elections) Seed Users Auxiliary Users 52 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 53. Conclusions of RQ 3 ● RQ3.1 - Affected users generate and consume tweets reflecting their pro-BJP ideology - part of an echo chamber ● RQ3.2 - Affected user group are more likely to belong to a heavily pro-BJP state ● RQ3.3 - Possibility of auxiliary users to hold relations with members of BJP party , whereas seed users are likely to be supporters or lower-level worker of BJP and associated groups 53 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 54. RQ 4 - What is the reachability of NaMo content on Twitter in comparison to non-NaMo content?
  • 55. Correlation graphs normalized by followers count 55 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 56. CDF Graphs 56 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 57. Conclusions of RQ 4 ● Content posted through the NaMo App may not have much reachability. Thus, does not affecting the larger discourse on Twitter. ● Observed similar results for CAA dataset 57 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 58. Conclusions ● Less influence of NaMo App on Twitter - not affecting the larger discourse and shows its inability to change the narrative on Twitter ● Association of auxiliary users with members of BJP party. ● Seed users (likely to be BJP workers) are responsible for cross posting content, were seen praising BJP party and demeaning other political parties. ● Affected users are part of online echo chambers - unable to break into a wider and more diverse audience ● Limited reachability of the NaMo content on Twitter 58 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 59. Future Work ● To identify users that BJP should target to increase its reach or influence through NaMo App ● A deeper content analysis to identify topics which are often cross posted ● If exists in future, one can incorporate party apps and its data to compare their visibility 59 CSE, IIT Kanpur Influence of NaMo App on Twitter MTech Thesis Presentation
  • 60. 60 Limitations and Challenges: ● Naive Political Classification ○ Considered only hashtags, and not the context of hashtags ■ Sarcasm tweets/Parity users - intent of tweet not aligning with intent of involved hashtags - may get allocated a wrong political leaning ○ Classification was done only for binary (BJP vs Others) on the basis of top 600 annotated hashtags - while this could have also been done in multi class problem (Pro-BJP, Anti-BJP, Pro-Cong, Anti-Cong or other political party) ● Users with more neutral hashtags and less political hashtags - will get assigned a leaning on the basis of political hashtags only ● To get a better understanding of count of affected users- we could also compare these values by considering the tweets not from lok sabha elections posted by users from these states ● Code-Mixed Data