meta ds interview | meta vo proxy | Product case

1,125Views

Interviewing for a Data Science role at Meta isn’t just about crunching numbers — it’s a mix of technical depth and product thinking. You’llusually face a blend of SQL, stats, and experiment design questions, but what really stands out is how much they care about product sense. Instead of onlytesting if you can code, they want to see if you can use data to actually drive decisions. It’s tough, but also one of the more interestinginterview processes out there.

meta ds interview | meta vo proxy | Product case

面試將分為兩部分（共約 30-45 分鐘）：

第 1 部分：案例分析（10-20 分鐘）
第二部分：技術分析（10-20 分鐘）

對於此面試的技術部分，您可以回答下列問題 SQL, Python, or R請使用您覺得最舒適的語言；我們沒有偏好！請注意，我們有幾個資料處理的問題是設計在 SQL (以表格和列作為輸入），但我們接受任何可以達到解決方案的程式語言。

Meta DS VO Detail

1. Introduce yourselves to each other

The first step is a self introduction part, Later, we entered the technical interview.

2. SQL – video call

Please provide the following two tables and write SQL statements. This question is not difficult, but it is not too difficult

table 1
callerid | recipientid | ds | call_id | duration

Question 1: How many users have started a call with more than 3 people in the last 7 days?

SELECT COUNT(DISTINCT callerid) AS user_count
FROM table1
WHERE ds >= DATE_SUB(CURRENT_DATE, INTERVAL 7 DAY)
GROUP BY callerid
HAVING COUNT(DISTINCT recipientid) > 3;

Question 2: What percentage of daily active users from France (‘fr’) were on a video call yesterday?

WITH daily_active_users AS (
    SELECT user_id
    FROM table2
    WHERE ds = DATE_SUB(CURRENT_DATE, INTERVAL 1 DAY)
      AND country = 'fr'
      AND dau_flag = 1
),
video_call_users AS (
    SELECT DISTINCT callerid AS user_id
    FROM table1
    WHERE ds = DATE_SUB(CURRENT_DATE, INTERVAL 1 DAY)
    UNION
    SELECT DISTINCT recipientid AS user_id
    FROM table1
    WHERE ds = DATE_SUB(CURRENT_DATE, INTERVAL 1 DAY)
)
SELECT
    COUNT(DISTINCT v.user_id) / COUNT(DISTINCT d.user_id) * 100 AS percentage_video_call_users
FROM daily_active_users d
LEFT JOIN video_call_users v ON d.user_id = v.user_id;

3.Product Case

There are three types of notifications:

Time critical (your friend is on live)
Feedback (ask what’s your opinion on something)
Security notification (password change)

Question 1: How to define the good quality of notification and what data do you need?

I mentioned the goal, and the goal is improvement engagement. Then, I discussed some high-level success metrics, such as DAU/MAU, time spent on the app,etc. Afterward, I talked about driver metrics, which are more actionable metrics, such as click-through rate (CTR) and whether users complete theirintended actions after clicking. Finally, I covered some counter metrics, like opt-out rate and uninstall rate.

Question 2: How do you check if, let’s say, 35% CTR is good or not?

I mentioned that different types of notifications would have different benchmarks. Security-related notifications might have higher benchmarks.This section primarily involves interview content related to Product Case, focusing on the quality evaluation and measurement metrics ofnotifications. The core topics discussed include:

How to define a high-quality notification? (Evaluating from multiple dimensions, including engagement, success rate, CTR, opt-out rate, etc.)

How to determine if a 35% CTR is a good metric? (It should be compared against benchmarks for different notification types.)