How do you build people you may know recommendation for LinkedIn?

Q 

Approach

1. Understand the prompt better

2. Establish the business goal 

3. Outline the attributes for the recommendation engine

4. Prioritize the key attributes

Understand the prompt

Candidate: Do we suggest friends recommendation for all users or only for users with appropiate privacy settings?

Interviewer: All users, independent of privacy setting 

Candidate: Can we use the data tracked outside the facebook platorm for making the recommendation?

Interviewer: Yes

3. Do we build the recommendation only for app or for both desktop and app

Interviewer: both

Business goal 

I would build the recommendation with the aim to facilitate the quality connection on facebook that aligns with facebook's mission of making the world a more connected. 

Recommendation engine framework

I would identify the set of attributes and weigh them based on their importance and build a composite score and select the top profiles for recommendation in terms of the composite score.

Attributes that the recommendation engine would be based-off of

1. Contacts: Facebook can access the contacts details of the user and suggest their contacts for friends

2. Communities: Facebook can make friends suggestion based on the shared communities such as school, college, employers as well as based on the shared facebook groups

3. Location: It is convienient to form a online frienship and continue it offline if the people are in close proximity. Hence location is another helpful signal 

4. Mutual friends: Facebook can leverage the social graph and recommend connections between users with large of mutual friends. People with large number of mutual friends tend to enjoy each other company. 

5. Common interests: Based on the people likes, comments, shares, we can identity the people with common interest in movies, sports, celebrities, music, food, political orientation, religious beliefs, etc. and levarge this information to make recommendations

6. Ethinicity/Race: People from same race tend to share similar worldview, culture, food, music, lifestyle; it can help to build a quality relationship

7 . Profession: It is a another signal that could lead to a potential quality relationship

8. Generation/Age group: People that belong to the same generation tend to share same world outlook and they can enjoy each others company well.

9. Gender: Women in general tend to comfortably send a friend request to someone of same gender than otherwise. So it is another useful signal.

10. Recent activity: In order to facility the conversation between people, those people must be active on the platform. Therefore, we want to recommend the profile that are recently active.

Prioritization:

Key attributes to based the recommendation engine off of, in that order

1. Saved contacts: It is strongest signal for developing the social bond because saved contacts suggest that these people are already connected

2. Mutual friends: More the mutual friend, higher the likelihood that user will send the request to the suggest friends, and other person accept the request

3. Communities: Institutions, workplace, sports team are factors based on which people can easily develop social bonds.

4. Recent activity: More activity the user, more likely that user will accept interest quickly

5. Ethnicity: It is another strong indicator for developing strong social conection.