FB 2
FB 2
Problem Statement
How would you prevent hate, misinformation, or deep-fakes on Facebook?
Clarifying Questions
1. What is considered hate speech on Facebook?
Hate speech includes content that promotes violence, discrimination, or harm based on
race, religion, gender, or other protected groups.
2. Will the solution cover only user-generated content, or should we also include ads,
comments, and sponsored content?
Initially, we will focus on user-generated content, but we may expand to cover ads,
comments, and sponsored content in the future.
3. Should users help find harmful content?
Yes, users can report harmful content.
4. What actions should be taken on flagged content?
Flagged content will be reviewed by human moderators who will decide whether to
remove, hide, or label the content.
5. Is this tool meant to be globally applicable or customized for specific regions?
The tool should be globally applicable but should have the flexibility to accommodate
regional regulations and cultural sensitivities in the future.
6. How do we measure success?
Success will be measured by reducing harmful content, increasing moderation
efficiency, and improving user trust through transparency.
7. What is the timeline for developing and releasing the first version of this tool?
The goal is to release an MVP within the next 4-5 months, focused on flagging harmful
content with a focus on the most critical issues.
8. Are there any budgetary or technical constraints we should consider?
No budget constraint. The solution needs to integrate seamlessly with Facebook’s
existing systems while being scalable.
9. How will cultural differences in hate speech be addressed globally?
We will consider local cultures and laws, using regional experts to guide moderation.
This ensures content is reviewed based on each region’s specific needs.
User Segmentation
1. Regular Users
These users frequently browse and interact with content on Facebook. They might engage with
posts from friends, family, and public pages.
Needs: They need a smooth, safe browsing experience, free from harmful content like hate
speech and misinformation. They want to know that the platform is actively working to protect
them from dangerous content.
Pain Points: They are frustrated by seeing harmful or misleading content in their feeds and want
a quick way to report or avoid such posts.
Needs: They need tools that help them create content responsibly. They also require the ability
to manage and moderate comments or content that may lead to misinformation or hate
speech.
Pain Points: They struggle with ensuring their content is not misinterpreted or manipulated
(deep-fakes) and want better control over how their content is shared or flagged.
Needs: They need efficient tools for flagging harmful content, verifying facts, and reviewing
flagged posts to maintain the integrity of information.
Pain Points: They often face an overwhelming amount of flagged content, especially during
high-stakes events like elections, and need support from AI tools to process and review content
quickly.
User Persona
Name Anjali Sharma
Age 28
Background Anjali is a young professional who works in marketing and spends time on
Facebook to stay connected with her friends, family, and the latest news. She
is active in various groups and often follows updates from public pages.
However, she is frustrated by harmful content like hate speech and
misinformation. She wants to enjoy the platform without feeling concerned
about the accuracy of what she sees.
Tech Moderate – She uses Facebook on her phone regularly and is familiar with the
Savviness platform’s features but doesn’t have a deep understanding of the back-end
moderation systems or AI tools that detect harmful content.
• Difficulty in identifying harmful content: She struggles to differentiate between real news
and fake/misleading content, especially when deep-fakes or manipulated images/videos
are involved.
• Reporting challenges: When she does encounter harmful content, Anjali finds it frustrating
to report it efficiently. The reporting tools aren’t quick or clear enough, and she wonders if
her report will actually lead to action.
• Feeling unsafe: The presence of harmful content, including hate speech and
misinformation, leaves her feeling unsafe on the platform, questioning the quality and
accuracy of everything she sees.
• Limited transparency: After reporting harmful content, Anjali has no way of knowing if her
report was processed or if the content was actually removed. This lack of feedback makes
her feel powerless in curbing harmful content.
Core Problem
The core problem is that Anjali, along with other users, struggles to enjoy a safe, informative,
and engaging experience on Facebook due to the prevalence of harmful content like hate
speech, misinformation, and deep-fakes. The tools currently available for reporting and dealing
with such content are ineffective, cumbersome, and lack transparency, which makes users feel
unsafe and powerless.
Pain Points
1. Exposure to Harmful Content
Users frequently encounter harmful content like hate speech, misinformation, and deep-
fakes in their feeds.
Impact: This creates an unsafe browsing experience and makes users feel uncomfortable
and distrustful of the platform.
Impact: Users become confused and may unknowingly share false information, which
undermines trust in the platform.
Impact: Users feel frustrated and powerless, as they can't quickly address harmful content
they come across.
Impact: This lack of transparency reduces trust in the platform’s ability to effectively handle
harmful content, leaving users feeling ignored.
Impact: This increases anxiety and makes users question the accuracy of information,
diminishing their confidence in the platform’s ability to maintain a safe environment.
Impact: This leads to a sense of helplessness, causing users to lose trust in the platform’s
commitment to tackling harmful content.
Business Alignment
The business alignment focuses on improving user experience, trust, compliance, and
advertiser relationships, all of which are essential for Facebook’s long-term success. By
focusing on preventing hate, misinformation, and deep-fakes, Facebook can build a safer and
more reliable platform, ensuring the business can continue to grow sustainably while meeting
both user and regulatory expectations.
These three pain points should be prioritized as they have the highest RICE scores and will have
the most significant impact on improving user experience and trust on the platform.
Proposed Solutions
Solution 1: Enhanced AI-Based Detection & Real-Time Flagging
Pain point: Exposure to Harmful Content
Description:
Improved AI models will be deployed to detect harmful content such as hate speech,
misinformation, and deepfakes across text, images, and videos. This system will use machine
learning and natural language processing to analyze user content in real-time and automatically
flag inappropriate posts. The flagged content will then be reviewed by Facebook moderators or
fact-checkers for further action.
How It Works:
• If a post is flagged, the system sends alerts to Facebook's moderation team or third-
party fact-checkers for verification.
Features Required:
• Advanced Text & Image Analysis: Deep learning models for analyzing text and images
for harmful content.
• Real-time Flagging System: Instant detection and flagging of harmful content for quick
action.
• Content Review System: Moderators and automated systems review flagged content
for final actions.
Impact:
This solution will reduce harmful content exposure, enhancing users' safety and trust on the
platform. It ensures quicker identification and removal of hate speech, misinformation, and
deepfakes, fostering a healthier environment.
Description:
This solution involves prominently labeling content flagged as misinformation, deepfakes, or
potentially harmful. Facebook will partner with credible third-party fact-checkers to validate
content and provide context. These labels will give users easy-to-understand warnings about
the accuracy of the content they encounter.
How It Works:
• Content flagged as potentially harmful or misleading is marked with clear labels (e.g.,
“Fact-Checked,” “Possible Misinformation,” “Deepfake Detected”).
• Fact-checkers verify content, and once validated, a “Fact-Checked” label with a source
link is placed.
• Users can click on the label to access more context or references supporting the
accuracy of the post.
• Deepfake detection tools can be used to flag altered videos or images with a warning.
• Content Labeling System: A feature to apply and display accurate content labels like
"Fact-Checked" or "Deepfake Detected."
• User Interface for Context: A clickable label that leads to more information, such as
articles or sources, that explain the label.
• Deepfake Detection Tools: AI tools that analyze videos and images for signs of
manipulation, such as mismatched voices or unnatural movements.
Impact:
By clearly labeling harmful content, users will be empowered to make informed decisions. This
solution increases trust in the platform, reduces misinformation spread, and helps users easily
identify and avoid false content.
Description:
This solution aims to provide users with visibility into the content moderation process by giving
them access to a Content Moderation Dashboard. After users report harmful content, they will
be able to track the status of their reports in real time, including updates on actions taken (e.g.,
removal of content, warnings to users). This transparency will help users understand the
process and feel assured that their concerns are being addressed.
How It Works:
• After reporting content, users can visit their dashboard to track the status of their report
(e.g., "Under Review," "Content Removed," "No Action Taken").
• Users will receive updates on the status of flagged content, including any action taken
or reasons for no action.
• When content is flagged, users can get detailed feedback explaining why content was
removed, flagged, or allowed to stay.
• Users will have access to a history of their reports, seeing how their concerns were
handled previously.
Features Required:
• User Dashboard: A simple, intuitive interface that displays all reported content and the
current status of each report.
• Status Indicators: Clear, real-time indicators (e.g., "Under Review," "Removed," "No
Action Taken") for reported content.
Impact:
Providing users with real-time updates on their reports builds trust and transparency. Users will
feel more in control, reducing frustration and improving confidence in the platform’s ability to
manage harmful content.
2. The AI system scans the content using machine learning and natural language
processing to detect harmful content (hate speech, misinformation, deepfakes).
5. The user receives a notification that their content has been flagged and is under review.
6. The moderators or fact-checkers review the flagged content and make a decision (e.g.,
removing, tagging, or allowing the post to remain live).
7. The user receives feedback about the status of their post (removed, flagged, or kept
live).
8. If the post is removed, the user gets an explanation regarding the reason (e.g., "Hate
speech detected" or "Misinformation detected").
2. Content is flagged as potentially harmful or misleading and marked with a label (e.g.,
“Fact-Checked,” “Possible Misinformation,” or “Deepfake Detected”).
4. Once verified, a “Fact-Checked” label is applied, along with a source link to validate the
claims.
5. The user clicks on the label to view more information, such as the context or references
supporting the claim.
7. The user clicks the label to understand how the content was detected as a deepfake.
8. The user decides whether to trust, share, or avoid the post based on the information
provided.
3. The user visits the Content Moderation Dashboard to track the status of their report.
4. The dashboard displays a list of reported content, with status indicators such as “Under
Review,” “Removed,” or “No Action Taken.”
5. The user checks the current status of their report and waits for updates.
6. Once the moderation process is complete, the status is updated to reflect the action
taken (content removed, flagged, or retained).
7. The user receives feedback on the actions taken, such as content removal or a warning
issued to the user who posted the content.
8. If the content is flagged or removed, detailed feedback is provided explaining why the
decision was made (e.g., "Hate speech" or "Misinformation").
9. The user can see their history of reported content and track how previous reports were
handled.
Metrics
Success Metrics
Metric Frequency Target Metric Explanation
90% of flagged
Tracks the speed of responding to
content acted
Real-Time flagged content. Quick action reduces
Weekly upon within 24
Moderation Rate the spread of harmful content and
hours (within 2
builds user trust.
months)
10 | P r o d u c t D e s i g n Muhamid Ali
Metric Frequency Target Metric Explanation
11 | P r o d u c t D e s i g n Muhamid Ali
Counter Metrics
Counter Metric Frequency Target Metric Explanation
Keep moderation-
A high volume of moderation-related
related support
Customer Support tickets indicates that users are facing
Weekly tickets under 5%
Ticket Volume issues with the system, whether it's
of total tickets
with flagging, reporting, or feedback.
(within 6 months)
12 | P r o d u c t D e s i g n Muhamid Ali
Phase Timeline Key Milestones Deliverables
1. High-fidelity UI/UX
1. Develop solution design for designs and
MVP (AI-based flagging system) wireframes
Phase 2: Design & Week 4 - 2. Finalize UI/UX designs and 2. Completed initial
Prototyping Week 6 user flows user testing results
3. Initial usability testing of 3. Feedback
prototypes with target users incorporated into
prototypes
13 | P r o d u c t D e s i g n Muhamid Ali
Phase Timeline Key Milestones Deliverables
1. Moderation
1. Public release of moderation transparency
transparency dashboard dashboard live
Phase 10: Transparency Week 43 - 2. Collect feedback and usage 2. Initial marketing
Dashboard Launch Week 46 data from users push for feature
3. Analyze and optimize based adoption
on feedback 3. Continuous
performance tracking
14 | P r o d u c t D e s i g n Muhamid Ali
Key Milestones
• Week 6: Finalized design and prototypes for AI-based flagging system.
• Week 12: AI-based flagging system MVP development complete.
• Week 20: Public launch of AI-based flagging system (MVP).
• Week 24: Iterations and optimizations based on user feedback after MVP launch.
• Week 32: Misinformation labeling system developed and integrated with fact-checkers.
• Week 36: Full rollout of misinformation and deepfake labeling system.
• Week 42: Moderation transparency dashboard development complete.
• Week 46: Public launch of moderation transparency dashboard.
• Week 52: Final product optimizations and successful delivery of all features.
15 | P r o d u c t D e s i g n Muhamid Ali