How do you go about determining the prevalence of fake news on Facebook and what impact it is having on the platform's users?

Question

Anonymous · Accepted Answer

Take a sample from the feed. Use human labels to mark whether the content was
fake or not. For a more detailed approach, build a ML model using signals like
source of content, link quality, vitality, comments and user feedback signals.
Use diff in diff to measure the impact on engagement.

Anonymous · Answer

Clarification - 
 * What is defined as fake news?
 * Why does it matter? 
 * Impact to what? User retention? user engagement? Acquisition rate?

Let's define fake news as posts from new users who create a large amount of
content and have low friend count & friend invites. Offer to define "new" &
"low" if interviewer wants you to spend time there. If this is a brand new
definition, start small and have human review to validate that definition is in
fact reasonable.

Let us consider only US created posts that appear in the newsfeed.

Let's say the impact we prioritize is user engagement. Few metric options:
 * login freq or 
 * total time spent on newsfeed or
 * Total time spent on FB
 * User activity i.e. reactions & comments

Pick one. 
Check if the selected metric is co-related to how many fake news posts appear in
their news feed? 
Segment this by user type (new vs established)

Anonymous · Answer

Look at how many posts in the news category gets tagged as spam or reported.
Also look into the virality, as fake news spreads faster. Impact could be
estimated by tracking bounce rates when this news is viewed or number of reports
as scam. also look at churn rate

Anonymous · Answer

* What isfake news:
 * News that have inaccurate/incorrect information
 * User/page generated, shared from other platform
 * Or created by anti social groups, extreme groups with agenda
   
 * Impact:hurt credibility of information
 * discourage user from engaging
 * churn in the end

* Identify fake news:
   Users reported fake news:by number of reports vs probability of fake
   Sample posts and do manual screening by agents
   
   For posts that do not have reports, find proxy: sensitive keywords, unknown
   source
   stratified sampling,
   Existing labels - build a classification model

How do you go about determining the prevalence of fake news on Facebook and what impact it is having on the platform's users?

Practice this question with AI

Go Premium

Practice More Questions

Community Answers

Unlock Community Insights

Unlock Community Insights

Unlock Community Insights

Unlock Community Insights

Master Data Science Interviews