My LinkedIn onsite had five rounds in total:
Data presentation
I was given two tables. One showed the current number of LinkedIn Learning courses in each category, and the other showed the number of courses and participants over time. I had to analyze them myself, then make slides and present.
SQL + Python
For SQL, I was given two experiment tables. One was a cohort table with users, their experiment entry times, and treatment/control assignment. The other contained a value for each user at each time. I had to define a metric myself and calculate the experiment's impact, mainly the difference in means, point estimate, and confidence interval.
Python was very easy, basically data manipulation like joins and groupbys. I've forgotten the exact questions.
Product
An RCA case: The number of job applications generated by job recommendation emails suddenly dropped one day. Analyze why. The main thing was to look at the funnel, from email impressions to opens, clicks, and applications, to see which step had gone wrong. The eventual cause was that the model's weights had changed, so it recommended more promoted jobs, which users didn't like.
Stats + Experimentation
The statistics question was the same queueing question from other interview reports. I was asked for the mean and variance of a Poisson distribution.
For experimentation, I was asked how to run an experiment when launching a new feature. One key point was not to artificially give it too much exposure at the start. First leave it in its natural placement to assess whether the feature itself works, then optimize exposure afterward.
HM
Basically a pure resume deep dive.
Discussion
Loading comments…