Cracking the Facebook Data Scientist Interview — Part 1
Introduction
There are a lot of resources on the web that explain the interview format for Data Scientist — Analytics position at Facebook. A good resource, quite a copy of what a recruiter would send to you as a candidate, is nicely elaborated here. However, a lot of candidates still feel a little lost while beginning their preparation, just like me. The purpose of this article is to help with the preparation style/ approach to problem solving and supporting content. Also, a quick look at the reasons to why Facebook, despite sharing so much material and interview details, still has a less than 1% intake conversion (from what was mentioned to me by the recruiters).
What more you need to know?
The Data Scientist role at Facebook is divided into 3 areas:
* Data Scientist — Engineer
* Data Scientist — Analytics
* Data Scientist — Core ML
So if you’re reviewing this article for anything aside of the Data Scientist- Analytics role, this might not be as helpful.
The Interview format:
Data Scientist — Analytics(DSA) interview has the following breakdown (most times):
Phone/Video screen: The initial technical screen consists of two parts (for a total of ~45 minutes)
- Product sense and analytics (10–20 minutes)
- Technical and Data Processing (10–20 minutes)
Onsite Interview: Again, this could be different based on the team but most likely the below structure is expected. This information is also available on public forums like Reddit/Glassdoor/Medium
- Analysis Case: Product Interpretation
- Analysis Case: Applied Data
- Quantitative Analysis
- Technical Analysis
You’ll also spend 1:1 time with a Data Scientist during your break to learn more about their life at Facebook. This is usually a 45-minute lunch interview that they’ll let you take a break or talk through what they work on at Facebook.
The Preparation
This is going to be the heart of my post, so let’s get into it. I am going to try and keep it very generic since we all have different strengths when it comes to the 4 pillars of Data Science: Math/Stats, Coding, ML theory, Productionizing models. I will take the top-down approach of explaining, starting with the high-level questions that allows us to think how the questions can be like.
Key Note: Interviewers at Facebook are extremely smart! What they really want to understand is “How you think?”, “How you approach a problem?”, complex math/coding can be learnt.
Product Interpretation: This round is similar to the Product Management interview, however it’s not exactly the same. A Data Scientist wears the Product Management hat (correctly so) to start solving a business problem - considering all possible cases to come up with different hypothesis and ways to prioritize and test them.
- Why do you think they made certain decisions about how it works?
- What could be done to improve the product?
- What kind of metrics you’d want to consider when solving for questions around health, growth, or the engagement of a product?
- How would you measure the success of different parts of the product?
- What metrics would you assess when trying to solve business problems related to our products?
- How would you tell if a product is performing well or not?
- How would you set up an experiment to evaluate any new products or improvements?
Let’s take an example: if I were to “Evaluate the News Feed” or any other feature, one approach could be as follows:
- Before jumping to an answer, pause and ASK clarifying questions or confirm your understanding of the question. I cannot stress this enough.
- Be equipped with a minimalistic structure of how you want to tackle this and talk to your interviewer about it. Eg. “I’d like to walk you through my understanding of the how the News Feed works, then discuss the goals and target audiences, key metrics and possibly some new features that might help us boost our metrics, and conclude our discussion. Does that sound ok to you?”. And maybe structure it out on the white board as well.
- How does the News Feed work? — You can talk in context of people who post (hereon referred to as producers) and for people who interact (hereon referred to as consumers) — like/share/comment/react.
- Pause, seek confirmation from your interviewer about your understanding.
- Once you confirm your understanding is good, define what the goal of a news feed is on a high level? Eg. Engagement that stems retention and why, increasing acquisition and why? and so on.
- Define what metrics would you use to track the goals. In order to come up with this, think about the possible user interactions. See below snapshot that will help you how to think about metrics. Prioritize your metrics. “I’d like to focus on user engagement so I think the top metrics I’d like to consider are # of interactions/unit time, avg # of interactions interaction per rolling 30 days, etc. I think these should be our most important metrics because of …”
- Are there any ways you can improve your product? (This is more of a case-by-case based discussion, be prepared with an answer but make sure the required questions are first answered). If the question itself is on improvement, then this should be in the conversation sooner than later.
- Summary
To assist with the preparation, I suggest the following prep guide, in order of priority:
Step 1: Understand the questions here: Glassdoor-1, Glassdoor-2, Glassdoor-3, Glassdoor-4, Glassdoor-5, Glassdoor-6
- Some additional questions:
- If 70% of Facebook users on iOS use Instagram, but only 35% of Facebook users on Android use Instagram, how would you investigate the discrepancy?
- FB would like to change the UI of the composer (the one users use to write a post) to be like INSTAGRAM (instead of a box ,they would they will add a “+” button at the bottom of the page), how would you test if this is a good idea
Step 2: See one of the answer styles: https://medium.com/stellarpeers/how-would-you-measure-the-success-of-facebook-stories-d4a520327119, https://medium.com/stellarpeers/how-would-you-measure-the-success-or-failure-of-a-product-feature-24f9d0b6f9e7
Step 3: Internalize the structure, DO NOT memorize it https://www.productmanagementexercises.com/how-to-answer-a-metrics-question
Step 4: Start with this problem here. They have a sample questions with a precise answer to what aspects need to be considered while thinking about a problem
Step 5: View the free videos in 1.5x — 2x mode. While watching these videos, try to first make notes of how YOU would answer them https://www.tryexponent.com/courses/pm
Step 6: MOST IMPORTANT: Talk about it! Discuss it with your friends. Give your friend a call and talk “Why the Facebook News feed was built” according to you, just do a back and forth conversation, you will internalize a pattern. Someone who is analytical can be super helpful, doesn’t have to be a Data Scientist.
Step 7: Learn about the Network Effect in A/B tests and understand how it can impact https://engineering.linkedin.com/blog/2019/06/detecting-interference--an-a-b-test-of-a-b-tests, GCP Network Effect problem and solution:http://www.unofficialgoogledatascience.com/2018/01/designing-ab-tests-in-collaboration.html
Backpocket information:
- Population of the US is 300 million.
- Life expectancy is 80 years.
- Number of people per household = 2.5
- How much shampoo does each person use per wash?
- What is a store’s mark-up per item?
- Android market: ~75%
- iOS market: ~23%
- Learn about the generations and Facebook target audiences: General Millennial Facts and their spending habits: https://www.lexingtonlaw.com/blog/credit-cards/millennial-spending-habits.html
- FB $$$ https://techcrunch.com/2019/10/30/facebook-earnings-q3-2019/
Please view the Math/Stats portion in my next article.
Happy to help :)
