[PDF] Understanding Information Weaponization via Data-driven Analysis





Previous PDF Next PDF



Paper Cyber-metapragmatics and alterity on reddit.com

creating subreddits commenting on threads—and their relationship to the “ For the “culture wars” taking place between Tumblr





The angry internet

large social media platforms Twitter



The Technological Factors of Reddit: Communication and Identity on

14 févr. 2016 Thus relational technology and networks focuses on the building and maintaining of relationships through the sharing of content. The term



Guidance on Research using Social Networking Sites

website and form relationships with other users of the same website who can access websites and apps Reddit



Exploring the Magnitude and Effects of Media Influence on Reddit

paper investigates these relationships on Reddit. ment study of 4chan's politically incorrect forum and its ef- fects on the web.



How recently emerged Reddit communities influenced our discourse

Reddit communities that participated in the Gamestop and AMC market turmoil have influenced relationships that go beyond discussing stocks.



Understanding Information Weaponization via Data-driven Analysis

Why Do We Care About 4Chan? Reddit. (six selected subreddits). 620K. 40K. 301K. 4chan (/pol/) ... Reveals causal relationships.



Exploring the Magnitude and Effects of Media Influence on Reddit

paper investigates these relationships on Reddit. ment study of 4chan's politically incorrect forum and its ef- fects on the web.



Analyzing Genetic Testing Discourse on the Web Through the Lens

relationship between health and social networks such as Twitter and Reddit to the best of our 2.3 Quantitative Studies of Twitter

Understanding Information Weaponization via Data-driven AnalysisEmiliano De Cristofaro emilianodc.com WARNINGCONTENT IN THIS TALK IS OFFENSIVEAND UNCENSORED2

Heard of 4chan?3

Heard of 4chan?3

Heard of 4chan?3

Heard of 4chan?3

Heard of 4chan?3

Heard of 4chan?3

Heard of 4chan?3

Heard of 4chan?3

Why Do We Care About 4Chan?4

Why Do We Care About 4Chan?4

Why Do We Care About 4Chan?4

Why Do We Care About 4Chan?4

What is 4chan?5

What is 4chan?An image-board forumOrganized in boards (70 at the moment)5

What is 4chan?An image-board forumOrganized in boards (70 at the moment)An "original poster" (OP) creates a new thread by making a postSingle image attached5

What is 4chan?An image-board forumOrganized in boards (70 at the moment)An "original poster" (OP) creates a new thread by making a postSingle image attachedOther users can reply:With or without images, possibly add references to previous posts, quote text, etc. 5

What is 4chan?An image-board forumOrganized in boards (70 at the moment)An "original poster" (OP) creates a new thread by making a postSingle image attachedOther users can reply:With or without images, possibly add references to previous posts, quote text, etc. 5

/pol/ - Politically Incorrect Board6 /pol/ - Politically Incorrect Board6 /pol/ - Politically Incorrect BoardExtremely lax moderationAlmost anything goes6

Anonymity & Ephemerality7

Anonymity & EphemeralityUsers do not need to register an account to participateAnonymity is the default (and preferred) behavior

7

Anonymity & EphemeralityUsers do not need to register an account to participateAnonymity is the default (and preferred) behavior

"Some" degree of permanence and identifiability is supportedCan enter a name along with their posts (no authentication though)

7

Anonymity & EphemeralityUsers do not need to register an account to participateAnonymity is the default (and preferred) behavior

"Some" degree of permanence and identifiability is supportedCan enter a name along with their posts (no authentication though)

Threads get "archived" after a whileActually all posts deleted after a week7

Challenges of Measuring 4chan

(and other fringe communities)8

Challenges of Measuring 4chan

(and other fringe communities)1.Not your typical social network (anonymous/ephemeral)8

Challenges of Measuring 4chan

(and other fringe communities)1.Not your typical social network (anonymous/ephemeral)2.Their actions not limited to 4chan, need to look at other

platforms to understand their impact8

Challenges of Measuring 4chan

(and other fringe communities)1.Not your typical social network (anonymous/ephemeral)2.Their actions not limited to 4chan, need to look at other

platforms to understand their impact3.Knowing what they're talking about is not easy8

Challenges of Measuring 4chan

(and other fringe communities)1.Not your typical social network (anonymous/ephemeral)2.Their actions not limited to 4chan, need to look at other

platforms to understand their impact3.Knowing what they're talking about is not easy4.You might get attacked, doxxed, etc.8

Datasets/pol//sp//int/TotalThreads217K14.4K24.9K256KPosts8.3M1.2M1.4M10.9MJune 30 to September 12, 20169

DatasetsMethodology:Visit the "catalog"Snapshot every 5 minsOnce a thread is archived, retrieve full/final contents from 4plebs.org/pol//sp//int/TotalThreads217K14.4K24.9K256KPosts8.3M1.2M1.4M10.9MJune 30 to September 12, 20169

DatasetsMethodology:Visit the "catalog"Snapshot every 5 minsOnce a thread is archived, retrieve full/final contents from 4plebs.orgWe're still crawling.../pol//sp//int/TotalThreads217K14.4K24.9K256KPosts8.3M1.2M1.4M10.9MJune 30 to September 12, 20169

Hate Speech?Crowdsourced dictionaryManually filtered a bit/pol/ by far most hate speech use/pol/ 12%/sp/ 7.3%/int/ 6.3%Twitter 2.2%or?10

Raids on other platforms (e.g., YouTube)11

Raids on other platforms (e.g., YouTube)Someone posts a YouTube linkMaybe with a prompt like "you know what to do"11

Raids on other platforms (e.g., YouTube)Someone posts a YouTube linkMaybe with a prompt like "you know what to do"Thread is an aggregation point for raidersE.g., "Hah! I called that person a n******!"11

Raids on other platforms (e.g., YouTube)Someone posts a YouTube linkMaybe with a prompt like "you know what to do"Thread is an aggregation point for raidersE.g., "Hah! I called that person a n******!"If raid is taking place:Peak in YouTube comments while thread alive?/pol/ thread and YT comments synchronized?11

Raids on other platforms (e.g., YouTube)12

Raids on other platforms (e.g., YouTube)12

Activity Peaks14% of videos see peak commenting activity during /pol/ thread lifetimeYT videos with peaks during 4chan threadDetermined via PDF of commenting timeseries13

Synchronization14

SynchronizationTwo series, second randomly shifted from first by 0.2s on avg14

Synchronization01-101Sample Lag (s)Two series, second randomly shifted from first by 0.2s on avgBlue lines ! per-sample lagRed area ! density of the lagsPeak of density curve = 0.2s14

Validation-2.50.02.50.0000.0050.0100.0150.020Hate comments per secondSynchronization Lag (105 seconds)YT comments have hateYT comments have no hate15

In fact... can we predict?16

In fact... can we predict?16Yes, we can!

Ensemble of classifiers determines the likelihood that a video will be raided with AUC up to 94%

What About Cyberbullying?70.6% of young people say they have seen bullying in their schools 9% of students in grades 6-12 experienced cyberbullying 15% of high school students (grades 9-12) were electronically bullied in the past year17

Cyberaggression vs Cyberbullying18

Cyberaggression vs CyberbullyingCyberbullying: intentionally aggressive behavior, repeated over time, which involves an imbalance of power18

Cyberaggression vs CyberbullyingCyberbullying: intentionally aggressive behavior, repeated over time, which involves an imbalance of powerCyberaggression: purposefully saying or doing something to hurt someone18

Cyberaggression vs CyberbullyingCyberbullying: intentionally aggressive behavior, repeated over time, which involves an imbalance of powerCyberaggression: purposefully saying or doing something to hurt someoneCan we detect it?18

Building a ClassifierCrawling data from Twitter from June to August 2016Baseline: 1M random tweetsHate-related: 650K tweets based on 309 bully- and hate-related hashtags#GamerGate + 308 co-appeared ones(GamerGate was a coordinated campaign of harassment in the online gaming world, revolving around sentiments of sexism, feminism, and "social justice")Pre-processing: removing spam, stop words, punctuation marks, etc.19

Crowdsourced labeling20

Ground Truth & Features21

Ground Truth & Features9,484 tweets, 1,307 users (4.5% bullies, 3.4% aggressors, 31.8% spammers, 60.3% normal)21

Ground Truth & Features9,484 tweets, 1,307 users (4.5% bullies, 3.4% aggressors, 31.8% spammers, 60.3% normal)User Features: avg. #posts, account age, #subscribed lists, verified account, posts' interarrival time, default profile image21

Ground Truth & Features9,484 tweets, 1,307 users (4.5% bullies, 3.4% aggressors, 31.8% spammers, 60.3% normal)User Features: avg. #posts, account age, #subscribed lists, verified account, posts' interarrival time, default profile imageText Features: #hashtags, #uppercases, #emoticons, #URLs, sentiment, avg. word embedding score, hate and curse scores21

Ground Truth & Features9,484 tweets, 1,307 users (4.5% bullies, 3.4% aggressors, 31.8% spammers, 60.3% normal)User Features: avg. #posts, account age, #subscribed lists, verified account, posts' interarrival time, default profile imageText Features: #hashtags, #uppercases, #emoticons, #URLs, sentiment, avg. word embedding score, hate and curse scoresNetwork Features: popularity (#follower, #friends), reciprocity, avg. power difference with mentioned users, hubs and authority, influence (eigenvector centrality, closeness centrality), communities (clustering coefficient, louvain modularity)21

22

Results23

How do communities influence each other?24

25

4chan ! Twitter25

4chan ! Twitter25

26
26

Reddit ! Twitter26

The Pizzagate Conspiracy Theory27

The Pizzagate Conspiracy TheoryData Provider27

The Pizzagate Conspiracy TheoryData ProviderTheory Generator27

The Pizzagate Conspiracy TheoryData ProviderTheory GeneratorTheory Incubators & Gateway to mainstream "world"27

The Pizzagate Conspiracy TheoryData ProviderTheory GeneratorTheory Incubators & Gateway to mainstream "world"Large-scale Disseminator27

Idea...28

Idea...Study the appearance of alternative and mainstream news URLs within the platforms28

Idea...Study the appearance of alternative and mainstream news URLs within the platformsBuild a sequence of appearance for each URL according to the timestamps28

Idea...Study the appearance of alternative and mainstream news URLs within the platformsBuild a sequence of appearance for each URL according to the timestamps28

Idea...Study the appearance of alternative and mainstream news URLs within the platformsBuild a sequence of appearance for each URL according to the timestampsBuild a graph with the sequences28

The Data99 mainstream and alternative ("fake") news sourcesPlatformPosts/CommentsAlternative URLsMainstream URLsTwitter486K42K236KReddit (six selected subreddits)620K40K301K4chan (/pol/)90K9K40K29

Here's What The Graph Looks LikeMainstream News SourcesAlternative News Sources30

Quantify Influence? Hawkes Processes!31

Quantify Influence? Hawkes Processes!Assume K processes Each with a rate of events (i.e., posting of a URL), called the background rate31

Quantify Influence? Hawkes Processes!Assume K processes Each with a rate of events (i.e., posting of a URL), called the background rateAn event can cause impulse responses in other processesIncreases the rates of other processes for a period of time31

Quantify Influence? Hawkes Processes!Assume K processes Each with a rate of events (i.e., posting of a URL), called the background rateAn event can cause impulse responses in other processesIncreases the rates of other processes for a period of timeEnables us to be confident about the number of events caused by another event on the source process (weight)Reveals causal relationships31

Hawkes Processes ExampleX

Hawkes Processes ExampleRedditX

Hawkes Processes ExampleRedditTwitterX

Hawkes Processes ExampleRedditTwitter/pol/X

Hawkes Processes ExampleRedditTwitter/pol/1X

Hawkes Processes ExampleRedditTwitter/pol/1X

Hawkes Processes ExampleRedditTwitter/pol/1X

Hawkes Processes ExampleRedditTwitter/pol/12X

Hawkes Processes ExampleRedditTwitter/pol/12X

Hawkes Processes ExampleRedditTwitter/pol/12X

Hawkes Processes ExampleRedditTwitter/pol/123X

Hawkes Processes ExampleRedditTwitter/pol/123X

Hawkes Processes ExampleRedditTwitter/pol/123X

Hawkes Processes ExampleRedditTwitter/pol/123X

Hawkes Processes ExampleRedditTwitter/pol/1234X

Hawkes Processes ExampleRedditTwitter/pol/1234X

Hawkes Processes ExampleRedditTwitter/pol/1234X

Hawkes Processes ExampleRedditTwitter/pol/12345X

Hawkes Processes ExampleRedditTwitter/pol/123456X

Hawkes Processes ExampleRedditTwitter/pol/1234567X

For Our Purposes...Hawkes model with 8 processesOne for each platform/communityDistinct model for each URLFit each model with Gibbs samplingCalculate the percentage of events created because of events happened in each of the other processes32

How Communities Influence Each Other?Twitter top influencers for alternative URLs o/r/The_Donald (2.72%)o/pol/ (1.96%)o/r/politics (1.1%)Twitter top influencers for mainstream URLso/r/politics (4.29%)o/pol/ (3.01%)o/r/The_Donald (2.97%)

33

How Communities Influence Each Other?Twitter top influencers for alternative URLs o/r/The_Donald (2.72%)o/pol/ (1.96%)o/r/politics (1.1%)Twitter top influencers for mainstream URLso/r/politics (4.29%)o/pol/ (3.01%)o/r/The_Donald (2.97%)

33

How Communities Influence Each Other?Twitter top influencers for alternative URLs o/r/The_Donald (2.72%)o/pol/ (1.96%)o/r/politics (1.1%)Twitter top influencers for mainstream URLso/r/politics (4.29%)o/pol/ (3.01%)o/r/The_Donald (2.97%)

These seemingly tiny Web communities can really punch above their weight class when it comes to influencing the greater Web33

Memes are fun! 34

Memes are fun! 34

35
35
35
35

Not always though...35

Memes in politics36

Memes in politics36

Memes in politicsMemes have become a popular, and seemingly effective, method to transmit ideology.Memes have been weaponized37

Memes processing pipeline38

Memes processing pipeline38

Datasets# of posts1.4B1.0B48M12M15K# of posts with images242M62M13M955K15K# of Images114M40M4M235K706K# of unique pHashes74M30M3M193K597K39

Top memes per Web community40

Top memes per Web community40

Top memes per Web community40

Top memes per Web community40

Top memes per Web community40

Who are the most popular people in memes?41

42
How are memes shared over time?Political MemesRacist Memes43 How are memes shared over time?Political MemesRacist Memes44 How are memes shared over time?Political MemesRacist Memes2nd US presidential debate44

How are memes shared over time?Political MemesRacist Memes2016 US elections2nd US presidential debate44

How are memes shared over time?Political MemesRacist Memes2016 US electionsGab activity increase 20172nd US presidential debate44

How are memes shared over time?Political MemesRacist Memes2016 US electionsGab activity increase 2017/pol/ constant shareGab activity increase in 20172nd US presidential debate44

Communities' influence (racist memes)/pol/ is most influential in terms of spreading racist memes45

Communities' efficiency (racist memes)If we look at theinfluence normalizedto the number ofmemes posted, theThe_Donald is mostefficient in terms ofdisseminating memes46

What about state-level adversaries?47

What about state-level adversaries?47

Research Questions•How do state-sponsored actors operate and evolve? •How does the behavior of state-sponsored trolls compare to random users?

•More importantly, what was their influence on the Web with respect to the dissemination of news?48

What are they posing as?Nudging users to follow themNews accounts?Trump supporters49

Where are they allegedly posting from?50

What hashtags they shared?51

Do Russian trolls change their screen names?9% of the Russian troll accounts changed their screen nameUp to 4 times per accountE.g., from "OnlineHouston" to "HoustonTopNews"Clear attempt to pose as local news outletIn our baseline dataset 19% of the accounts changed their screen nameUp to 11 times per account52

Do Russian trolls delete their tweets?53

When did Russian trolls deleted their tweets?One month before 2016 US elections54

Influence1K trolls caused 2.6% of all Russian state-sponsored news outlets URLs (i.e., RT) on Twitter's 1%.1K trolls caused 0.6% of all other news outlets URLs on Twitter's 1%.55

Main FindingsWe find differences in the use of the Twitter platform between trolls and random users Trolls seem to reset their "personas" by changing names and deleting tweets Particularly influential in spreading Russian state-sponsored URLs on Twitter and other platforms56

In conclusion, we developed methodologies...57

In conclusion, we developed methodologies...Data CollectionCrawling 4chan, Reddit, Gab, Twitter, YouTubeProcessing, cleaning, storing, indexing, etc.57

In conclusion, we developed methodologies...Data CollectionCrawling 4chan, Reddit, Gab, Twitter, YouTubeProcessing, cleaning, storing, indexing, etc.Data AnalysisStatistical tools to study distributions, behaviors, etc.Longitudinal analysis, understanding trends and change pointsInfluence estimationImage analysisClustering, contextual analysis, unsupervised learningLanguage modeling, topic extractionClassification and prediction[...]Exploratory & hypotheses-driven57

In conclusion, we developed methodologies...Data CollectionCrawling 4chan, Reddit, Gab, Twitter, YouTubeProcessing, cleaning, storing, indexing, etc.Data AnalysisStatistical tools to study distributions, behaviors, etc.Longitudinal analysis, understanding trends and change pointsInfluence estimationImage analysisClustering, contextual analysis, unsupervised learningLanguage modeling, topic extractionClassification and prediction[...]Exploratory & hypotheses-drivenData SharingEverything open sourceEasy to use, facilitate qualitative/social science research57

Moving Forward58

Moving ForwardThere are still many problems remaining in the broad space58

Moving ForwardThere are still many problems remaining in the broad spaceWe have made a lot of progress into gaining a quantitative understanding58

Moving ForwardThere are still many problems remaining in the broad spaceWe have made a lot of progress into gaining a quantitative understandingWe now need to deploy proactive systems in productionReal time detection of harassment and online attacksActively fight weaponized informationPolicyAccountability58

Thank you!59

quotesdbs_dbs7.pdfusesText_13
[PDF] 4chan reddit spacing

[PDF] 4change energy billing address

[PDF] 4change energy budget saver 12

[PDF] 4change energy cancellation

[PDF] 4change energy customer reviews

[PDF] 4change energy customer service

[PDF] 4change energy deposit

[PDF] 4change energy google reviews

[PDF] 4change energy houston reviews

[PDF] 4change energy login

[PDF] 4change energy maxx saver 12

[PDF] 4change energy number

[PDF] 4change energy phone number

[PDF] 4change energy plan

[PDF] 4change energy reviews bbb