SlideShare a Scribd company logo
1 of 31
GAME UX
SUMMIT ’17
#GAMEUXSUMMIT ‘17 / TORONTO
Toxicity and Moderation:
Data-Based Approaches at Twitch
Ruth Toner
Data Scientist, Twitch Interactive
Introduction
TWITCH:
o Live streaming and on-demand video
o Fourth largest source of internet traffic in the US, mostly (but not only!)
gaming content
 In a single month:
o 2.2 Million broadcasters and content creators, including gamers, esports, devs,
and non-gaming content
o 15 Million Daily Active Viewers
o 1billion+ chat and private messages sent
1/28
Twitch Chat– Why Do We Care?
2/28
Chat:
- Main way users
interact with
broadcaster
- Subscriptions and
“cheering” (tips)
- Key part of funnel to
engaged, paying
viewers
- We want to make
being social on
Twitch a good
experience
Introduction
1 BILLION MESSAGES = HARASSMENT AND
ABUSE HAPPEN
 This talk = how Twitch uses data to understand:
o How abuse happens on Twitch
o How we build better tools to fight it
o How can we combine data science and human insight?
3/28
Human-centric Data Science
 Intelligence Augmentation: “The ultimate goal is not building machines that think
like humans, but designing machines that help humans think better.”
 Guszcza(1), Lewis, Evans-Greenwood “Cognitive collaboration: Why humans and computers think better
together” Deloitte University Press Jan 2017
4/28
Smaller scale
insights
The Sweet Spot
Good Data Science
+ UX
Pure data, but
also “Artificial
Stupidity”1
Pure Qualitative Pure Quantitative
Moderation + Data Science
1. Extent
How do we describe +
quantify abuse on
Twitch?
2. Impact
How do we answer
questions about the
impact of abuse and
our tools?
3. Tools
How do we use data to
build effective tools to
fight abuse?
5/28
The Goal:
• Help our content creators can build the communities they want
(within limits…)
• No one leaves Twitch because they feel unsafe or harassed
1. Extent
FIRST, WE NEED TO
UNDERSTAND OUR
DATA…
6/28
Understanding our data
7/28Twitch Chat
8/28Any User: Twitch Site-wide Moderation
o Reports are sent
from a user to
Twitch’s site-wide
Human Admin
moderation staff
o These admins can
issue a Strike: a
temporary
suspension or
permanent ban
from Twitch
Data Source: Reports and Strikes
 Safety Violation Signal: TWITCH TERMS OF SERVICE VIOLATIONS
 TOS: Among many other things, basic rules of conduct for broadcasting and chatting
(no harassment, threats, impersonation, etc.)
 A viewer or broadcaster is reported for violating the basic rules of conduct governing
behavior on Twitch, and can receive a strike limiting use of their account.
 Human Judgement:
 Reports: People mislabel spam as harassment. Behavior was bad but didn’t break
ToS. People report each other as a joke.
 Strikes: 100% accurate source of data, but not a complete picture of unsafe
behavior.
9/28
10/28Channel Moderators: Timeouts and Bans
Every channels can appoint
moderators who can:
o Time Out chatters
(temporary)
o Ban chatters (permanent)
Data Source: Timeouts and Bans
 Safety Violation: COMMUNITY RULE BREAKING
 A channel moderator can ban or time-out someone from participating
from chat when they break the rules of a community
We give broadcasters autonomy to decide what conversation is
acceptable in their community (within Terms of Service limits…).
 Human Judgement: Not all rule violations are safety violations.
Moderators also moderate for spam, for links or all-caps, for spoilers, or
(again!) as a joke (“Mods plz ban me!”).
11/28
12/28
Moderator:
Troll:
Troll:
Broadcaster: AutoMod
Data Source: AutoMod
 Safety Violation: UNACCEPTABLE LANGUAGE
 Broadcaster decides how ’risky’ they want language to be on their
channel, from just removing hate speech to forbidding cursing.
 Two Signals:
AutoMod ratings: how risky AutoMod thinks a chat message is.
Mod approvals + denials: what the channel moderators thought.
 Human Judgement: Missing social context for the messages.
13/28
Data from Moderation Tools
 Each Data Source: How safe or happy our viewers or broadcasters feel on Twitch
 BUT ALSO: False Positives, Noise, Unclear Signals
 “A flag is not merely a technical feature: It is a complex interplay between users and
platforms, humans and algorithms, and the social norms and regulatory structures of
social media.”
 Crawford and Gillespie, “What Is A Flag For? Social Media Reporting Tools and the Vocabulary of Complaint” New
Media & Society July 2014
 We understand these signals and noise by exploring data and talking to our users
14/28
Example: Two Types of Abuser
Question: What does a troll look like?
 Chatters suspended for harassment share a few things in
common:
 Multiple channel bans
 Younger than average accounts
 Higher than expected language risk
 However, if we talk to our admins and then take a closer look at
our data, it turns out this question is too simple…
15/28
Account Age:
Regular vs Suspended User
Example: Two Types of Abuser
Better Question: What do different types of
troll look like?
 We see two major subcategories!
 Chat Harassers: Higher risk language, young and old accounts
alike.
 Ban Evader: Younger accounts with low activity and levels of
verification.
 We need different solutions for different types of abuse
 Mixing quantitative analysis and qualitative assessment allowed
us to update our intuition about trolling…
16/28
(Suspended) Account Age:
Ban Evader vs Harasser
Abuse: Impact
NEXT, WE NEED TO
ASK THE RIGHT
QUESTIONS WITH THE
RIGHT TOOLS…
17/28
Measuring impact
Understanding our data
Data Science Tools: Questions + Problems
 We want to turn our qualitative user insights into testable hypotheses.
 A/B testing: Causal analysis, but ethical considerations + confusion…
 Better for smaller product iterations or helper tools.
 Quasi-experimental studies: Cheaper, but self selection effects +
confounding variables everywhere!
 Example: A channel which bans a lot of users may actually be a healthier
channel, since they have a staff of moderators and bots.
18/28
Viewership Impacts?
 Key Question: How does abusive behaviors impact
the health of our community?
 Reduced Broadcaster RETENTION?
 Reduced viewer ENGAGEMENT?
 Lots of 3rd party UX and DS research:
 Pew 2017 Research – Online Harassment
 Riot Games and other industry research
 Talking directly to our viewers and broadcasters
 Tanya DePass: “How to Keep Safe In the
Land of Twitch”
https://www.twitch.tv/videos/174334243
19/28
https://www.polygon.com/2012/10/17/3515178/the-league-of-
legends-team-of-scientists-trying-to-cure-toxic
Moderation Workload Impact?
 Key Question: What is it like to actually use our moderation
products?
 How fast can administrators respond to reports?
 How many actions do our human channel moderators need to perform when
they moderate a chat room?
 What are the gaps in the system?
 Start by talking to our user base and performing qualitative studies to
identify these pain points, and then try to study and verify them with our
quantitative data.
20/28
Growth and Moderation Workload
 User complaint:
 As chat gets bigger and faster,
have to mod faster and a
larger % of messages
 Very busy chats = have a full
moderation staff, but
moderation efficiency goes
down
 Solution: Build moderation tools
which reduce the amount of work
which our moderators need to do
per message.
21/28
Mod Action / Message: Extra Human Mod Staff:
Moderation Efficiency vs Conversation Speed:
Chat Message/Min Chat Message/Min1 msg
100 min
10 msg
1 second
Impact Study: Chat Rules
 Intended impact: Get rid of of timeouts and bans
caused by misunderstanding of channel rules.
 A/B Test: When entering a channel for the first time,
chatters were shown control and variant:
 Chat rules: click to agree
 No chat rules
 Results: No significant impact on chat participation, and
a statistically significant reduction in timeouts and bans
for the ‘click to agree’ variant!
22/28
GOG.com’s Twitch chat rules
Toxicity: Tools
LET’S USE THESE
LEARNINGS TO BUILD
SOMETHING THAT MAKES
OUR USERS SAFER
23/28
Intervention
Measuring impact
Understanding our data
AutoMod
 Data Product Problem: Can we help broadcasters
passively filter hate speech, bullying, and sexual
language they don’t want on their chat?
 Solution: AutoMod - automated filtering of language,
based on topic category and perceived level of risk
 Algorithm designed using a combination of statistical
learning and human qualitative review
24/28
Designing AutoMod
 Start with a pre-trained off-the-shelf ML solution
 Segments and normalizes each chat segment.
 Categorizes sentence fragments by risk topic (hate, sex, bullying, etc.) and severity
(high risk, medium risk, etc.)
 Can handle over ten languages, combos of words and emotes, misspellings, and
(important!) attempts to get around the filter.
25/28
Example:
Original: “Omg. You should killll yooorseeeeeefff.”
Parsed: [ omg ] [ {you/he/she} | should | {self harm} ]
no risk Bullying – High Risk Level
Designing AutoMod
 Making this work for Twitch:
 Compare, for sentence fragment f:
 Use Lf to flag individual expressions which were obvious false
positives or incorrectly rated.
 Chose risk thresholds for our preset options, Rule Levels 1-4
 Get it running in the field
 Initial dry run: DNC/RNC Conventions 2016
 Small closed beta to refine usability and filter accuracy.
26/28
𝐿 𝑓 ~ log
𝑁𝑓,𝑏𝑎𝑛𝑛𝑒𝑑 + 1
𝑁 𝑎𝑙𝑙,𝑏𝑎𝑛𝑛𝑒𝑑 + 1
𝑁𝑓,𝑛𝑜 𝑏𝑎𝑛 + 1
𝑁 𝑎𝑙𝑙,𝑛𝑜 𝑏𝑎𝑛 + 1
For fragment ‘f’ (and message counts Ncat):
AutoMod Risk Likelihood Lf of User Being
Banned for That Fragment
versus
Maintaining AutoMod
 Full opt-in launch of AutoMod on Dec15, 2016
 Improving Accuracy: Use Approve and Deny actions to
determine what AutoMod recommendations our users
agree and disagree with.
 L’f Factor: Surface list of recommended rule changes,
which are then vetted by our admin staff.
 Sep 2017: False positives reduced by 33% since launch!
 25% of all chat messages go through AutoMod
 Continue to develop based on performance and user
feedback...
27/28
𝐿′ 𝑓 ~ log
𝐶𝑓,𝑑𝑒𝑛𝑖𝑒𝑑 + 1
𝐶 𝑎𝑙𝑙,𝑑𝑒𝑛𝑖𝑒𝑑 + 1
𝐶𝑓,𝑎𝑝𝑝𝑟𝑜𝑣𝑒𝑑 + 1
𝐶 𝑎𝑙𝑙,𝑎𝑝𝑝𝑟𝑜𝑣𝑒𝑑 + 1
For fragment ‘f’ (and total unique
channels Ccat):
Conclusions
 Our Punchline: Quantitative analysis and qualitative research alone can’t capture
exactly what’s happening with safety in our products and community.
 Combine data science with qualitative learnings from our UX team, our admins, and
from talking to our viewers and broadcasters for better decisions
 Where we apply this:
 Extent: Figure out what signal your data is giving you about safety.
 Impact: What are the right questions we should be asking, and using what tools and
metrics?
 Tools: Using these data and questions, we can craft powerful tools for safety!
28/28
29
‘Kappa - Bob Ross Portrait’
By: twitch.tv/sohlol
Twitch TOS – Relevant Sections
 9. Prohibited Conduct
 You agree that you will comply with these Terms of Service and Twitch’s Community
Guidelines and will not:
 i. create, upload, transmit, distribute, or store any content that is inaccurate, unlawful,
infringing, defamatory, obscene, pornographic, invasive of privacy or publicity rights,
harassing, threatening, abusive, inflammatory, or otherwise objectionable;
 ii. impersonate any person or entity, falsely claim an affiliation with any person or entity, or
access the Twitch Services accounts of others without permission, forge another person’s
digital signature, misrepresent the source, identity, or content of information transmitted via
the Twitch Services, or perform any other similar fraudulent activity;
 v. defame, harass, abuse, threaten or defraud users of the Twitch Services, or collect, or
attempt to collect, personal information about users or third parties without their consent;
30

More Related Content

Similar to Moderation Tools and User Safety: Data-Driven Approaches at Twitch

Strategy for Mobile Keyboard
Strategy for Mobile KeyboardStrategy for Mobile Keyboard
Strategy for Mobile KeyboardIvano Digital
 
Auto Response System for Legal Consultation
Auto Response System for Legal ConsultationAuto Response System for Legal Consultation
Auto Response System for Legal Consultationijtsrd
 
DP1_160430723010_Divya.pptx
DP1_160430723010_Divya.pptxDP1_160430723010_Divya.pptx
DP1_160430723010_Divya.pptxDivyaPatel729457
 
Essay On Visit To Taj Mahal In Hindi Language
Essay On Visit To Taj Mahal In Hindi LanguageEssay On Visit To Taj Mahal In Hindi Language
Essay On Visit To Taj Mahal In Hindi LanguageSharon Miller
 
Twitter Data Analysis
Twitter Data Analysis Twitter Data Analysis
Twitter Data Analysis Manan Gadhiya
 
Collaboration for Good Futures
Collaboration for Good FuturesCollaboration for Good Futures
Collaboration for Good FuturesMike Linksvayer
 
Assessment Worksheet Aligning Risks, Threats, and Vuln.docx
Assessment Worksheet Aligning Risks, Threats, and Vuln.docxAssessment Worksheet Aligning Risks, Threats, and Vuln.docx
Assessment Worksheet Aligning Risks, Threats, and Vuln.docxfestockton
 
IRJET- Improved Real-Time Twitter Sentiment Analysis using ML & Word2Vec
IRJET-  	  Improved Real-Time Twitter Sentiment Analysis using ML & Word2VecIRJET-  	  Improved Real-Time Twitter Sentiment Analysis using ML & Word2Vec
IRJET- Improved Real-Time Twitter Sentiment Analysis using ML & Word2VecIRJET Journal
 
digitaltech_university_0.rarDigitalTech Universitycontact.docx
digitaltech_university_0.rarDigitalTech Universitycontact.docxdigitaltech_university_0.rarDigitalTech Universitycontact.docx
digitaltech_university_0.rarDigitalTech Universitycontact.docxduketjoy27252
 
Pres dae20
Pres dae20Pres dae20
Pres dae20osimod
 
How To Write A Good TOK Essay A Simple Step-By-Step Guide
How To Write A Good TOK Essay A Simple Step-By-Step GuideHow To Write A Good TOK Essay A Simple Step-By-Step Guide
How To Write A Good TOK Essay A Simple Step-By-Step GuideAshley Smith
 
Trust influence and social media
Trust influence and social mediaTrust influence and social media
Trust influence and social mediaDawn Dawson
 
Can you trust everything?
Can you trust everything?Can you trust everything?
Can you trust everything?Colin Lieu
 
How To Write A Research Paper For Publication Outli
How To Write A Research Paper For Publication OutliHow To Write A Research Paper For Publication Outli
How To Write A Research Paper For Publication OutliDawn Henderson
 
Dbq Essay Outline Regents. Online assignment writing service.
Dbq Essay Outline Regents. Online assignment writing service.Dbq Essay Outline Regents. Online assignment writing service.
Dbq Essay Outline Regents. Online assignment writing service.Jennifer Subhedar
 
Star Writing Template Freebie By Cupcakes A
Star Writing Template Freebie By Cupcakes AStar Writing Template Freebie By Cupcakes A
Star Writing Template Freebie By Cupcakes AAudrey Britton
 
New 4Th Grade Opinion Essay Examples Pics - Exam
New 4Th Grade Opinion Essay Examples Pics - ExamNew 4Th Grade Opinion Essay Examples Pics - Exam
New 4Th Grade Opinion Essay Examples Pics - ExamAmanda Detwiler
 
Harvard Referencing Essay Example Topics And Wel
Harvard Referencing Essay Example Topics And WelHarvard Referencing Essay Example Topics And Wel
Harvard Referencing Essay Example Topics And WelJessica Myers
 
Knime social media_white_paper
Knime social media_white_paperKnime social media_white_paper
Knime social media_white_paperFiras Husseini
 

Similar to Moderation Tools and User Safety: Data-Driven Approaches at Twitch (20)

Strategy for Mobile Keyboard
Strategy for Mobile KeyboardStrategy for Mobile Keyboard
Strategy for Mobile Keyboard
 
Auto Response System for Legal Consultation
Auto Response System for Legal ConsultationAuto Response System for Legal Consultation
Auto Response System for Legal Consultation
 
DP1_160430723010_Divya.pptx
DP1_160430723010_Divya.pptxDP1_160430723010_Divya.pptx
DP1_160430723010_Divya.pptx
 
Essay On Visit To Taj Mahal In Hindi Language
Essay On Visit To Taj Mahal In Hindi LanguageEssay On Visit To Taj Mahal In Hindi Language
Essay On Visit To Taj Mahal In Hindi Language
 
Twitter Data Analysis
Twitter Data Analysis Twitter Data Analysis
Twitter Data Analysis
 
Collaboration for Good Futures
Collaboration for Good FuturesCollaboration for Good Futures
Collaboration for Good Futures
 
Assessment Worksheet Aligning Risks, Threats, and Vuln.docx
Assessment Worksheet Aligning Risks, Threats, and Vuln.docxAssessment Worksheet Aligning Risks, Threats, and Vuln.docx
Assessment Worksheet Aligning Risks, Threats, and Vuln.docx
 
IRJET- Improved Real-Time Twitter Sentiment Analysis using ML & Word2Vec
IRJET-  	  Improved Real-Time Twitter Sentiment Analysis using ML & Word2VecIRJET-  	  Improved Real-Time Twitter Sentiment Analysis using ML & Word2Vec
IRJET- Improved Real-Time Twitter Sentiment Analysis using ML & Word2Vec
 
digitaltech_university_0.rarDigitalTech Universitycontact.docx
digitaltech_university_0.rarDigitalTech Universitycontact.docxdigitaltech_university_0.rarDigitalTech Universitycontact.docx
digitaltech_university_0.rarDigitalTech Universitycontact.docx
 
Pres dae20
Pres dae20Pres dae20
Pres dae20
 
Consent Conference Göttingen July 2011
Consent Conference Göttingen July 2011Consent Conference Göttingen July 2011
Consent Conference Göttingen July 2011
 
How To Write A Good TOK Essay A Simple Step-By-Step Guide
How To Write A Good TOK Essay A Simple Step-By-Step GuideHow To Write A Good TOK Essay A Simple Step-By-Step Guide
How To Write A Good TOK Essay A Simple Step-By-Step Guide
 
Trust influence and social media
Trust influence and social mediaTrust influence and social media
Trust influence and social media
 
Can you trust everything?
Can you trust everything?Can you trust everything?
Can you trust everything?
 
How To Write A Research Paper For Publication Outli
How To Write A Research Paper For Publication OutliHow To Write A Research Paper For Publication Outli
How To Write A Research Paper For Publication Outli
 
Dbq Essay Outline Regents. Online assignment writing service.
Dbq Essay Outline Regents. Online assignment writing service.Dbq Essay Outline Regents. Online assignment writing service.
Dbq Essay Outline Regents. Online assignment writing service.
 
Star Writing Template Freebie By Cupcakes A
Star Writing Template Freebie By Cupcakes AStar Writing Template Freebie By Cupcakes A
Star Writing Template Freebie By Cupcakes A
 
New 4Th Grade Opinion Essay Examples Pics - Exam
New 4Th Grade Opinion Essay Examples Pics - ExamNew 4Th Grade Opinion Essay Examples Pics - Exam
New 4Th Grade Opinion Essay Examples Pics - Exam
 
Harvard Referencing Essay Example Topics And Wel
Harvard Referencing Essay Example Topics And WelHarvard Referencing Essay Example Topics And Wel
Harvard Referencing Essay Example Topics And Wel
 
Knime social media_white_paper
Knime social media_white_paperKnime social media_white_paper
Knime social media_white_paper
 

Recently uploaded

Vision and reflection on Mining Software Repositories research in 2024
Vision and reflection on Mining Software Repositories research in 2024Vision and reflection on Mining Software Repositories research in 2024
Vision and reflection on Mining Software Repositories research in 2024AyushiRastogi48
 
Microphone- characteristics,carbon microphone, dynamic microphone.pptx
Microphone- characteristics,carbon microphone, dynamic microphone.pptxMicrophone- characteristics,carbon microphone, dynamic microphone.pptx
Microphone- characteristics,carbon microphone, dynamic microphone.pptxpriyankatabhane
 
Call Girls in Munirka Delhi 💯Call Us 🔝9953322196🔝 💯Escort.
Call Girls in Munirka Delhi 💯Call Us 🔝9953322196🔝 💯Escort.Call Girls in Munirka Delhi 💯Call Us 🔝9953322196🔝 💯Escort.
Call Girls in Munirka Delhi 💯Call Us 🔝9953322196🔝 💯Escort.aasikanpl
 
Evidences of Evolution General Biology 2
Evidences of Evolution General Biology 2Evidences of Evolution General Biology 2
Evidences of Evolution General Biology 2John Carlo Rollon
 
zoogeography of pakistan.pptx fauna of Pakistan
zoogeography of pakistan.pptx fauna of Pakistanzoogeography of pakistan.pptx fauna of Pakistan
zoogeography of pakistan.pptx fauna of Pakistanzohaibmir069
 
TOTAL CHOLESTEROL (lipid profile test).pptx
TOTAL CHOLESTEROL (lipid profile test).pptxTOTAL CHOLESTEROL (lipid profile test).pptx
TOTAL CHOLESTEROL (lipid profile test).pptxdharshini369nike
 
‏‏VIRUS - 123455555555555555555555555555555555555555
‏‏VIRUS -  123455555555555555555555555555555555555555‏‏VIRUS -  123455555555555555555555555555555555555555
‏‏VIRUS - 123455555555555555555555555555555555555555kikilily0909
 
Gas_Laws_powerpoint_notes.ppt for grade 10
Gas_Laws_powerpoint_notes.ppt for grade 10Gas_Laws_powerpoint_notes.ppt for grade 10
Gas_Laws_powerpoint_notes.ppt for grade 10ROLANARIBATO3
 
Dashanga agada a formulation of Agada tantra dealt in 3 Rd year bams agada tanta
Dashanga agada a formulation of Agada tantra dealt in 3 Rd year bams agada tantaDashanga agada a formulation of Agada tantra dealt in 3 Rd year bams agada tanta
Dashanga agada a formulation of Agada tantra dealt in 3 Rd year bams agada tantaPraksha3
 
Recombinant DNA technology( Transgenic plant and animal)
Recombinant DNA technology( Transgenic plant and animal)Recombinant DNA technology( Transgenic plant and animal)
Recombinant DNA technology( Transgenic plant and animal)DHURKADEVIBASKAR
 
insect anatomy and insect body wall and their physiology
insect anatomy and insect body wall and their  physiologyinsect anatomy and insect body wall and their  physiology
insect anatomy and insect body wall and their physiologyDrAnita Sharma
 
Is RISC-V ready for HPC workload? Maybe?
Is RISC-V ready for HPC workload? Maybe?Is RISC-V ready for HPC workload? Maybe?
Is RISC-V ready for HPC workload? Maybe?Patrick Diehl
 
Artificial Intelligence In Microbiology by Dr. Prince C P
Artificial Intelligence In Microbiology by Dr. Prince C PArtificial Intelligence In Microbiology by Dr. Prince C P
Artificial Intelligence In Microbiology by Dr. Prince C PPRINCE C P
 
Temporomandibular joint Muscles of Mastication
Temporomandibular joint Muscles of MasticationTemporomandibular joint Muscles of Mastication
Temporomandibular joint Muscles of Masticationvidulajaib
 
BIOETHICS IN RECOMBINANT DNA TECHNOLOGY.
BIOETHICS IN RECOMBINANT DNA TECHNOLOGY.BIOETHICS IN RECOMBINANT DNA TECHNOLOGY.
BIOETHICS IN RECOMBINANT DNA TECHNOLOGY.PraveenaKalaiselvan1
 
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptxSOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptxkessiyaTpeter
 
Call Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCR
Call Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCRCall Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCR
Call Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCRlizamodels9
 

Recently uploaded (20)

Engler and Prantl system of classification in plant taxonomy
Engler and Prantl system of classification in plant taxonomyEngler and Prantl system of classification in plant taxonomy
Engler and Prantl system of classification in plant taxonomy
 
Vision and reflection on Mining Software Repositories research in 2024
Vision and reflection on Mining Software Repositories research in 2024Vision and reflection on Mining Software Repositories research in 2024
Vision and reflection on Mining Software Repositories research in 2024
 
Microphone- characteristics,carbon microphone, dynamic microphone.pptx
Microphone- characteristics,carbon microphone, dynamic microphone.pptxMicrophone- characteristics,carbon microphone, dynamic microphone.pptx
Microphone- characteristics,carbon microphone, dynamic microphone.pptx
 
Call Girls in Munirka Delhi 💯Call Us 🔝9953322196🔝 💯Escort.
Call Girls in Munirka Delhi 💯Call Us 🔝9953322196🔝 💯Escort.Call Girls in Munirka Delhi 💯Call Us 🔝9953322196🔝 💯Escort.
Call Girls in Munirka Delhi 💯Call Us 🔝9953322196🔝 💯Escort.
 
Evidences of Evolution General Biology 2
Evidences of Evolution General Biology 2Evidences of Evolution General Biology 2
Evidences of Evolution General Biology 2
 
zoogeography of pakistan.pptx fauna of Pakistan
zoogeography of pakistan.pptx fauna of Pakistanzoogeography of pakistan.pptx fauna of Pakistan
zoogeography of pakistan.pptx fauna of Pakistan
 
Hot Sexy call girls in Moti Nagar,🔝 9953056974 🔝 escort Service
Hot Sexy call girls in  Moti Nagar,🔝 9953056974 🔝 escort ServiceHot Sexy call girls in  Moti Nagar,🔝 9953056974 🔝 escort Service
Hot Sexy call girls in Moti Nagar,🔝 9953056974 🔝 escort Service
 
TOTAL CHOLESTEROL (lipid profile test).pptx
TOTAL CHOLESTEROL (lipid profile test).pptxTOTAL CHOLESTEROL (lipid profile test).pptx
TOTAL CHOLESTEROL (lipid profile test).pptx
 
‏‏VIRUS - 123455555555555555555555555555555555555555
‏‏VIRUS -  123455555555555555555555555555555555555555‏‏VIRUS -  123455555555555555555555555555555555555555
‏‏VIRUS - 123455555555555555555555555555555555555555
 
Gas_Laws_powerpoint_notes.ppt for grade 10
Gas_Laws_powerpoint_notes.ppt for grade 10Gas_Laws_powerpoint_notes.ppt for grade 10
Gas_Laws_powerpoint_notes.ppt for grade 10
 
Volatile Oils Pharmacognosy And Phytochemistry -I
Volatile Oils Pharmacognosy And Phytochemistry -IVolatile Oils Pharmacognosy And Phytochemistry -I
Volatile Oils Pharmacognosy And Phytochemistry -I
 
Dashanga agada a formulation of Agada tantra dealt in 3 Rd year bams agada tanta
Dashanga agada a formulation of Agada tantra dealt in 3 Rd year bams agada tantaDashanga agada a formulation of Agada tantra dealt in 3 Rd year bams agada tanta
Dashanga agada a formulation of Agada tantra dealt in 3 Rd year bams agada tanta
 
Recombinant DNA technology( Transgenic plant and animal)
Recombinant DNA technology( Transgenic plant and animal)Recombinant DNA technology( Transgenic plant and animal)
Recombinant DNA technology( Transgenic plant and animal)
 
insect anatomy and insect body wall and their physiology
insect anatomy and insect body wall and their  physiologyinsect anatomy and insect body wall and their  physiology
insect anatomy and insect body wall and their physiology
 
Is RISC-V ready for HPC workload? Maybe?
Is RISC-V ready for HPC workload? Maybe?Is RISC-V ready for HPC workload? Maybe?
Is RISC-V ready for HPC workload? Maybe?
 
Artificial Intelligence In Microbiology by Dr. Prince C P
Artificial Intelligence In Microbiology by Dr. Prince C PArtificial Intelligence In Microbiology by Dr. Prince C P
Artificial Intelligence In Microbiology by Dr. Prince C P
 
Temporomandibular joint Muscles of Mastication
Temporomandibular joint Muscles of MasticationTemporomandibular joint Muscles of Mastication
Temporomandibular joint Muscles of Mastication
 
BIOETHICS IN RECOMBINANT DNA TECHNOLOGY.
BIOETHICS IN RECOMBINANT DNA TECHNOLOGY.BIOETHICS IN RECOMBINANT DNA TECHNOLOGY.
BIOETHICS IN RECOMBINANT DNA TECHNOLOGY.
 
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptxSOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
 
Call Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCR
Call Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCRCall Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCR
Call Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCR
 

Moderation Tools and User Safety: Data-Driven Approaches at Twitch

  • 1. GAME UX SUMMIT ’17 #GAMEUXSUMMIT ‘17 / TORONTO Toxicity and Moderation: Data-Based Approaches at Twitch Ruth Toner Data Scientist, Twitch Interactive
  • 2. Introduction TWITCH: o Live streaming and on-demand video o Fourth largest source of internet traffic in the US, mostly (but not only!) gaming content  In a single month: o 2.2 Million broadcasters and content creators, including gamers, esports, devs, and non-gaming content o 15 Million Daily Active Viewers o 1billion+ chat and private messages sent 1/28
  • 3. Twitch Chat– Why Do We Care? 2/28 Chat: - Main way users interact with broadcaster - Subscriptions and “cheering” (tips) - Key part of funnel to engaged, paying viewers - We want to make being social on Twitch a good experience
  • 4. Introduction 1 BILLION MESSAGES = HARASSMENT AND ABUSE HAPPEN  This talk = how Twitch uses data to understand: o How abuse happens on Twitch o How we build better tools to fight it o How can we combine data science and human insight? 3/28
  • 5. Human-centric Data Science  Intelligence Augmentation: “The ultimate goal is not building machines that think like humans, but designing machines that help humans think better.”  Guszcza(1), Lewis, Evans-Greenwood “Cognitive collaboration: Why humans and computers think better together” Deloitte University Press Jan 2017 4/28 Smaller scale insights The Sweet Spot Good Data Science + UX Pure data, but also “Artificial Stupidity”1 Pure Qualitative Pure Quantitative
  • 6. Moderation + Data Science 1. Extent How do we describe + quantify abuse on Twitch? 2. Impact How do we answer questions about the impact of abuse and our tools? 3. Tools How do we use data to build effective tools to fight abuse? 5/28 The Goal: • Help our content creators can build the communities they want (within limits…) • No one leaves Twitch because they feel unsafe or harassed
  • 7. 1. Extent FIRST, WE NEED TO UNDERSTAND OUR DATA… 6/28 Understanding our data
  • 9. 8/28Any User: Twitch Site-wide Moderation o Reports are sent from a user to Twitch’s site-wide Human Admin moderation staff o These admins can issue a Strike: a temporary suspension or permanent ban from Twitch
  • 10. Data Source: Reports and Strikes  Safety Violation Signal: TWITCH TERMS OF SERVICE VIOLATIONS  TOS: Among many other things, basic rules of conduct for broadcasting and chatting (no harassment, threats, impersonation, etc.)  A viewer or broadcaster is reported for violating the basic rules of conduct governing behavior on Twitch, and can receive a strike limiting use of their account.  Human Judgement:  Reports: People mislabel spam as harassment. Behavior was bad but didn’t break ToS. People report each other as a joke.  Strikes: 100% accurate source of data, but not a complete picture of unsafe behavior. 9/28
  • 11. 10/28Channel Moderators: Timeouts and Bans Every channels can appoint moderators who can: o Time Out chatters (temporary) o Ban chatters (permanent)
  • 12. Data Source: Timeouts and Bans  Safety Violation: COMMUNITY RULE BREAKING  A channel moderator can ban or time-out someone from participating from chat when they break the rules of a community We give broadcasters autonomy to decide what conversation is acceptable in their community (within Terms of Service limits…).  Human Judgement: Not all rule violations are safety violations. Moderators also moderate for spam, for links or all-caps, for spoilers, or (again!) as a joke (“Mods plz ban me!”). 11/28
  • 14. Data Source: AutoMod  Safety Violation: UNACCEPTABLE LANGUAGE  Broadcaster decides how ’risky’ they want language to be on their channel, from just removing hate speech to forbidding cursing.  Two Signals: AutoMod ratings: how risky AutoMod thinks a chat message is. Mod approvals + denials: what the channel moderators thought.  Human Judgement: Missing social context for the messages. 13/28
  • 15. Data from Moderation Tools  Each Data Source: How safe or happy our viewers or broadcasters feel on Twitch  BUT ALSO: False Positives, Noise, Unclear Signals  “A flag is not merely a technical feature: It is a complex interplay between users and platforms, humans and algorithms, and the social norms and regulatory structures of social media.”  Crawford and Gillespie, “What Is A Flag For? Social Media Reporting Tools and the Vocabulary of Complaint” New Media & Society July 2014  We understand these signals and noise by exploring data and talking to our users 14/28
  • 16. Example: Two Types of Abuser Question: What does a troll look like?  Chatters suspended for harassment share a few things in common:  Multiple channel bans  Younger than average accounts  Higher than expected language risk  However, if we talk to our admins and then take a closer look at our data, it turns out this question is too simple… 15/28 Account Age: Regular vs Suspended User
  • 17. Example: Two Types of Abuser Better Question: What do different types of troll look like?  We see two major subcategories!  Chat Harassers: Higher risk language, young and old accounts alike.  Ban Evader: Younger accounts with low activity and levels of verification.  We need different solutions for different types of abuse  Mixing quantitative analysis and qualitative assessment allowed us to update our intuition about trolling… 16/28 (Suspended) Account Age: Ban Evader vs Harasser
  • 18. Abuse: Impact NEXT, WE NEED TO ASK THE RIGHT QUESTIONS WITH THE RIGHT TOOLS… 17/28 Measuring impact Understanding our data
  • 19. Data Science Tools: Questions + Problems  We want to turn our qualitative user insights into testable hypotheses.  A/B testing: Causal analysis, but ethical considerations + confusion…  Better for smaller product iterations or helper tools.  Quasi-experimental studies: Cheaper, but self selection effects + confounding variables everywhere!  Example: A channel which bans a lot of users may actually be a healthier channel, since they have a staff of moderators and bots. 18/28
  • 20. Viewership Impacts?  Key Question: How does abusive behaviors impact the health of our community?  Reduced Broadcaster RETENTION?  Reduced viewer ENGAGEMENT?  Lots of 3rd party UX and DS research:  Pew 2017 Research – Online Harassment  Riot Games and other industry research  Talking directly to our viewers and broadcasters  Tanya DePass: “How to Keep Safe In the Land of Twitch” https://www.twitch.tv/videos/174334243 19/28 https://www.polygon.com/2012/10/17/3515178/the-league-of- legends-team-of-scientists-trying-to-cure-toxic
  • 21. Moderation Workload Impact?  Key Question: What is it like to actually use our moderation products?  How fast can administrators respond to reports?  How many actions do our human channel moderators need to perform when they moderate a chat room?  What are the gaps in the system?  Start by talking to our user base and performing qualitative studies to identify these pain points, and then try to study and verify them with our quantitative data. 20/28
  • 22. Growth and Moderation Workload  User complaint:  As chat gets bigger and faster, have to mod faster and a larger % of messages  Very busy chats = have a full moderation staff, but moderation efficiency goes down  Solution: Build moderation tools which reduce the amount of work which our moderators need to do per message. 21/28 Mod Action / Message: Extra Human Mod Staff: Moderation Efficiency vs Conversation Speed: Chat Message/Min Chat Message/Min1 msg 100 min 10 msg 1 second
  • 23. Impact Study: Chat Rules  Intended impact: Get rid of of timeouts and bans caused by misunderstanding of channel rules.  A/B Test: When entering a channel for the first time, chatters were shown control and variant:  Chat rules: click to agree  No chat rules  Results: No significant impact on chat participation, and a statistically significant reduction in timeouts and bans for the ‘click to agree’ variant! 22/28 GOG.com’s Twitch chat rules
  • 24. Toxicity: Tools LET’S USE THESE LEARNINGS TO BUILD SOMETHING THAT MAKES OUR USERS SAFER 23/28 Intervention Measuring impact Understanding our data
  • 25. AutoMod  Data Product Problem: Can we help broadcasters passively filter hate speech, bullying, and sexual language they don’t want on their chat?  Solution: AutoMod - automated filtering of language, based on topic category and perceived level of risk  Algorithm designed using a combination of statistical learning and human qualitative review 24/28
  • 26. Designing AutoMod  Start with a pre-trained off-the-shelf ML solution  Segments and normalizes each chat segment.  Categorizes sentence fragments by risk topic (hate, sex, bullying, etc.) and severity (high risk, medium risk, etc.)  Can handle over ten languages, combos of words and emotes, misspellings, and (important!) attempts to get around the filter. 25/28 Example: Original: “Omg. You should killll yooorseeeeeefff.” Parsed: [ omg ] [ {you/he/she} | should | {self harm} ] no risk Bullying – High Risk Level
  • 27. Designing AutoMod  Making this work for Twitch:  Compare, for sentence fragment f:  Use Lf to flag individual expressions which were obvious false positives or incorrectly rated.  Chose risk thresholds for our preset options, Rule Levels 1-4  Get it running in the field  Initial dry run: DNC/RNC Conventions 2016  Small closed beta to refine usability and filter accuracy. 26/28 𝐿 𝑓 ~ log 𝑁𝑓,𝑏𝑎𝑛𝑛𝑒𝑑 + 1 𝑁 𝑎𝑙𝑙,𝑏𝑎𝑛𝑛𝑒𝑑 + 1 𝑁𝑓,𝑛𝑜 𝑏𝑎𝑛 + 1 𝑁 𝑎𝑙𝑙,𝑛𝑜 𝑏𝑎𝑛 + 1 For fragment ‘f’ (and message counts Ncat): AutoMod Risk Likelihood Lf of User Being Banned for That Fragment versus
  • 28. Maintaining AutoMod  Full opt-in launch of AutoMod on Dec15, 2016  Improving Accuracy: Use Approve and Deny actions to determine what AutoMod recommendations our users agree and disagree with.  L’f Factor: Surface list of recommended rule changes, which are then vetted by our admin staff.  Sep 2017: False positives reduced by 33% since launch!  25% of all chat messages go through AutoMod  Continue to develop based on performance and user feedback... 27/28 𝐿′ 𝑓 ~ log 𝐶𝑓,𝑑𝑒𝑛𝑖𝑒𝑑 + 1 𝐶 𝑎𝑙𝑙,𝑑𝑒𝑛𝑖𝑒𝑑 + 1 𝐶𝑓,𝑎𝑝𝑝𝑟𝑜𝑣𝑒𝑑 + 1 𝐶 𝑎𝑙𝑙,𝑎𝑝𝑝𝑟𝑜𝑣𝑒𝑑 + 1 For fragment ‘f’ (and total unique channels Ccat):
  • 29. Conclusions  Our Punchline: Quantitative analysis and qualitative research alone can’t capture exactly what’s happening with safety in our products and community.  Combine data science with qualitative learnings from our UX team, our admins, and from talking to our viewers and broadcasters for better decisions  Where we apply this:  Extent: Figure out what signal your data is giving you about safety.  Impact: What are the right questions we should be asking, and using what tools and metrics?  Tools: Using these data and questions, we can craft powerful tools for safety! 28/28
  • 30. 29 ‘Kappa - Bob Ross Portrait’ By: twitch.tv/sohlol
  • 31. Twitch TOS – Relevant Sections  9. Prohibited Conduct  You agree that you will comply with these Terms of Service and Twitch’s Community Guidelines and will not:  i. create, upload, transmit, distribute, or store any content that is inaccurate, unlawful, infringing, defamatory, obscene, pornographic, invasive of privacy or publicity rights, harassing, threatening, abusive, inflammatory, or otherwise objectionable;  ii. impersonate any person or entity, falsely claim an affiliation with any person or entity, or access the Twitch Services accounts of others without permission, forge another person’s digital signature, misrepresent the source, identity, or content of information transmitted via the Twitch Services, or perform any other similar fraudulent activity;  v. defame, harass, abuse, threaten or defraud users of the Twitch Services, or collect, or attempt to collect, personal information about users or third parties without their consent; 30