This document proposes a solution to recommend social content to users based on what their friends have viewed, while addressing privacy and limited display space issues. It suggests ranking content viewed by a user's friends based on popularity metrics, then displaying the top-ranked content to the user once per day on their newsfeed or profile page. To protect privacy, it recommends anonymizing the recommendations by only attributing content to "one of your friends" rather than by a specific friend's name.
3. The Problem
• We want to recommend content to a user
which was viewed by his friends recently.
• But we can’t recommend a user everything his
friends viewed because we have limited
display space.
• And there is a privacy issue because a user’s
friends might not want their viewing history
exposed.
5. The Solution
• Rank the slideshows viewed recently by a
user’s friends using some metric.
• Just show the top ranked slideshow from that
list.
• Display one slideshow per day to the user.
6. But what will be the ranking metric?
• Suppose a user has 100 friends,
• Each friend viewed 20 slideshows in the last
week.
• That gives us 100 X 20 = 2000 slideshows
• Rank them according to the total number of
views/favorites/comments the slideshow got.
7. How will we know the correctness
of the metric?
By measuring the CTR.
8. But where will we display these
recommendations?
In the newsfeed/profile page
10. But what is anonymizing ?
• It means hiding personal data about an entity.
• Show the heading “Slideshow S Viewed by
one of your friends”
• Never show “Slideshow S Viewed by your
friend Mr. X”
• Because we don’t want to make Mr. X upset.