This was presented by Alisa Nguyen, Annie Franco and Samidh Chakrabarti from Facebook at the Impacts of Civic Technology Conference (TICTeC 2018) in Lisbon on 18th April 2018. You can find out more information about the conference here: http://tictec.mysociety.org/2018
2. Overview
• Why We Work On Civic Discourse
• What Defines Healthy Conversation
• Four Case Studies for Improving Civic Conversations
• Q&A
3. Civic engagement is one of five major pillars
of how FB seeks to realize its mission
https://www.facebook.com/notes/mark-zuckerberg/building-gobal-community/10154544292806634/
Giving people a voice is a principle our community
has been committed to since we began.
As we look ahead to building the social infrastructure for a
global community, we will work on building new tools that
encourage thoughtful civic engagement.
Only through dramatically greater engagement can
we ensure [that] political processes reflect our values.
-- Mark Zuckerberg, February 2017
8. Overview
• Why We Work On Civic Discourse
• What Defines Healthy Conversation
• Four Case Studies for Improving Civic Conversations
• Q&A
9. We ask people through surveys and focus groups what
kind of conversation they find to be the most valuable
Rapid Feedback Surveys
▪ Triggeredbyclickonpager
(someoneactivelyreadingcomments)
▪ Translatedintotop14languages.
▪ Onequestionsurvey=>highresponserate!
Concept Questio
10. What kind of civic discourse do people want on Facebook?
Safe
Authentic
Meaningful
Effective
Civility
(e.g., Respectfulness)
Real People
Community
Acknowledgement
(i.e., Feeling Heard)
Offensiveness
(e.g., Hate Speech)
Misinformation
Divisiveness
Irrelevance
(i.e., Off Topic)
MORE… LESS…
11. Overview
• Why We Work On Civic Discourse
• What Defines Healthy Conversation
• Four Case Studies for Improving Civic Conversations
• Q&A
12. Overview
• Why We Work On Civic Discourse
• What Defines Healthy Conversation
• Four Case Studies for Improving Civic Conversations
• Safety: Kenya 2017 Election
• Q&A
13. How can we keep online speech from inciting real violence?
16. Advances in ML allowed us to shift to proactive reporting
1.
Train ML classifiers
for Kenya-specific slurs
2.
Use AI to proactively
flag suspect content
3.
Review manually by
market-specific teams
Proactive method was 60% more accurate
17. Overview
• Why We Work On Civic Discourse
• What Defines Healthy Conversation
• Four Case Studies for Improving Civic Conversations
• Authenticity: Ad Transparency
• Q&A
18. How to make sure people involved in civic
conversations really are who they say they are?
• US 2016 election showed
danger of inauthentic civic
content
• We’re constantly improving
our ability to catch fake
accounts
• Now stopping 1M every day at
the point of registration
• But how to prevent inauthentic
accounts run by real people?
Credit: Patch
19. Transparency: You’ll soon be able to go to any FB Page and
view all the ads they are currently running to anyone
20. Verification: All political advertisers will also be
required to verify their identity and their location
• Political ads from unverified
advertisers will be blocked
• Includes all kinds of political ads
(i.e., electoral + social/issue ads)
• Identity verification will be
similar to opening a bank acct
• Government Photo ID
• Social Security Number
• Location will be verified by
sending a code over postal mail
Note: Preliminary design. Continuing to iterate prior to launch.
21. Disclaimers:
Political ads will
be required to
show people who
paid for them
Note: Preliminary design. Continuing to iterate prior to launch.
22. Archive: Political ads will be in a searchable archive for 4 yrs
• Political Ad Archive will show…
• Full ad creative
• Amount spent
• Demographic info about ad’s
audience (age, gender, geo)
• Publicly Searchable
• Useful for watchdog groups
to investigate and flag ads
Note: Preliminary design. Continuing to iterate prior to launch.
23. Overview
• Why We Work On Civic Discourse
• What Defines Healthy Conversation
• Four Case Studies for Improving Civic Conversations
• Meaningfulness: Measuring Polarization
• Q&A
24. How can we ensure products don’t amplify polarization?
• “Bringing the world closer
together” is FB’s mission
• Responsibility to make sure
product changes don’t
exacerbate polarization
• To study this, we need to be able
to measure it first!
Credit: Brown University
25. Focus on Affective polarization, or ‘social distance’
↑↑
Feelings of Warmth
↑↓
Affective Polarization (the gap)
Out-group (people who disagree with you)
In-group (people who share your beliefs)
26. From TICTeC 2017: Our early attempt to measure
polarization via sentiment-based surveys
27. We need a better way of measuring affective polarization
What is the ideal metric?
• Real-time: Based on on-platform behaviors
• Privacy-conscious: Doesn’t rely on explicit political beliefs
• Valid: Correlates w/ people’s self-reported polarization
• Sensitive: Correlates with real-world trends
28. Preferential Liking is a FB analogue for affective polarization
People
Friends who engage
with similar news (↑↑)
p↑↑(like)
p↑↓ (like)
Friends who engage
with dissimilar news (↑↓)
• Key idea: measure behavioral positivity
gap
• We construct an ideology-agnostic
measure of similarity based on news
engagement patterns
• Preferential Liking:
“How much more likely are people to
‘like’ content from friends who engage
with similar news vs dissimilar news?”
29. Preferential Liking better correlates with survey measures
Likes of Content
from Similar Friends
Likes of Content from
Dissimilar Friends
Preferential Likes
(Delta of Above)
* ‘Similar’ is defined as friends who engage with similar news sources
30. Preferential Liking is sensitive to polarizing real-world events
Preferential Likes around Parkland Shooting
31. This is a first step to civic health metrics for tech products
Missing today from people’s product
dashboards: Civic health metrics!
Google Analytics Illustrative Example
32. Overview
• Why We Work On Civic Discourse
• What Defines Healthy Conversation
• Four Case Studies for Improving Civic Conversations
• Effectiveness: Constituent Badges
• Q&A
33. How can digital interactions w/ politicians be more effective?
Contacts
Responses
34. From TICTeC 2016: Politicians say they don’t respond to
people on FB because they don’t believe they are constituents
35. Constituent Badges let people show government officials they
live in their districts when interacting with their posts
36. By randomizing our rollout in two phases, we tried to
evaluate this feature’s impact on politician responses
Real World Outcome
Expected
Impact
Actual
Impact
Replies from politicians to constituent comments
(i.e., lift in reply rate amongst badge-eligible constituents)
Increase
37. Badges had no measurable effect on politician reply rate
Wave 1 (badge-eligible)
Wave 2 (not yet badge-eligible)
Comments Likes
38. Politicians still say they find badges to be valuable
and it may be effecting their legislative priorities
“Loving these constituent badges. I use them every week
with our Facebook Live Town Halls…. I moderate the
questions beforehand, and it’s been tremendously helpful
to be able to quickly determine which questions are from
constituents. It makes my job of finding questions much
much easier, and it helps the Senator be more able to
directly address his constituents during these Live events.
Loving it.”
~ A US Senator’s Office
39. Interestingly, among constituents, there is real change once
people opt-in to the badge! More sustained engagement…
Pre-Badge
Post-Badge
Per Day Rate Repeat Comment Rate
42. Badges shift mix of conversation towards people who are
traditionally under-represented in online civic discourse
43. While not increasing on-FB politician responsiveness,
Constituent Badges are empowering constituents
Real World Outcome
Expected
Impact
Actual
Impact
Replies from politicians to constituent comments
(i.e., lift in reply rate amongst badge-eligible constituents)
Increase Neutral
Repeated engagement by constituents - Increase
Substance of comments by constituents - Increase
Representation of women in comments - Increase
Sense of “Permission to Speak”!
44. Constituent Badges are proving popular and are improving the
way people and their representatives interact on FB
So far…
~ 7M badged comments created
~ 334M badged comments viewed
~ 23% opt-in rate (last 30d)
Stay tuned for mobile rollout!
45. Overview
• Why We Work On Civic Discourse
• What Defines Healthy Conversation
• Four Case Studies for Improving Civic Conversations
• Q&A
46. • This has been a tough year not just for FB, but for all of us
• The optimistic vision we’ve all had for civic tech feels at risk
• But there’s reason to be hopeful– evidence that tech can improve outcomes!
Conclusion: This is a challenging time, but be hopeful