LinkedIn Data Scientist Interview Experience — Product Sense, SQL, and a Pandas Group-By Slip

Interviewers: the main interviewer and a shadow interviewer, both very friendly.

Question 1: Product Sense — how would you analyze a drop in traffic from the Home Page to the Profile Page? (20 mins)

My approach:

• Start with a qualitative check: open LinkedIn's homepage myself to see if there's an obvious change or bug. For example, there was a recent user report that searching for someone's name showed "null null" — that kind of bug needs to be ruled out first.
• Collect macro data: has LinkedIn's total traffic changed? Has average time on site changed?
• Collect user path data: where do users go after landing on the homepage? Where do users who used to visit the Profile Page go now?

Interviewer follow-up: If total dwell time hasn't changed, what do you think is going on?

My answer: I suspected a product interaction change meant users no longer needed to click into the Profile page. For example, maybe the homepage added a hover feature — hovering over someone's name previews their basic info, so users can learn about the person without jumping to the Profile page. That would naturally lower Profile page visits, but it's a positive product improvement, not user loss.

Question 2: Product Sense — the homepage Feed changed from "everything" to "only relevant content." How would you measure success? (20 mins)

Measuring from the Viewer side: run an A/B test comparing the treatment group against the control group on engagement metrics — likes, comments, article click-through rate, etc.

Interviewer follow-up: Would the Poster side have concerns?

My answer: since content is now pushed to more precisely matched users, Posters should actually get higher-quality engagement. You could show Posters their content's total engagement volume, to help them see the value that relevance filtering brings.

Question 3: SQL (5 mins)

Join two tables (LEFT JOIN).

Question 4: Python (5 mins)

Calculate a certain ratio — not too hard overall, but partway through I got nervous and forgot to GROUP BY the first table before joining, which caused a small mess. Being nervous made my thinking a bit scattered.

Summary: the questions were medium difficulty and the interviewer was very friendly, giving a lot of guidance. For the Pandas part, remember to GROUP BY before you JOIN, or the result will come out wrong. The product questions were examined pretty carefully — you need to be able to connect a data anomaly back to a product feature change.