Pinterest Data Scientist 面试指南：Product Sense & SQL 是核心

Pinterest Data Scientist Interview Guide: Product Sense & SQL Focus

[!NOTE] Pinterest 的 Data Science 面试以 "Product-Driven" 著称。与 Google/Meta 不同，Pinterest 更看重你对 "User Discovery" 和 "Recommendation" 的业务理解。

本文基于 2026 最新 Data Scientist 面试真题，深度拆解 Pinterest 的考察重点。

Module 1: Product Sense (The Core)

Pinterest 的核心是 "Discovery Engine"。面试中，你需要像 Product Manager 一样思考。

• Understanding the Product:
  • Pinterest 不是 Social Media (like Instagram)，而是 Visual Search Engine。
  • Metric 重点不是 "Time Spent"，而是 "Successful Saves (Pins)" 和 "Click-through to Website"。
• Sample Question: "We want to launch a feature that allows users to shop directly from Pins. How would you measure success?"
  • North Star Metric: Gross Merchandise Value (GMV) attributed to Pins.
  • Guardrail Metric: User engagement (saves/clicks) on non-shoppable pins. (Don't hurt the core experience).

Module 2: SQL & Data Processing (Deep Dive)

Pinterest 的数据量极大 (Petabytes level)。SQL 考核非常硬核。

1. Schema Awareness

面试官通常会给你三个核心 Table： * users: user_id, country, signup_date * pins: pin_id, category, creator_id * actions: user_id, pin_id, action_type ('save', 'click', 'impression'), timestamp

2. Complex Query Example (Window Functions)

Question: "Find the top 3 categories for each user based on their 'save' actions in the last 7 days."

WITH user_category_counts AS (
    SELECT 
        a.user_id,
        p.category,
        COUNT(*) as save_count
    FROM actions a
    JOIN pins p ON a.pin_id = p.pin_id
    WHERE a.action_type = 'save'
      AND a.timestamp >= NOW() - INTERVAL '7 DAYS'
    GROUP BY 1, 2
),
ranked_categories AS (
    SELECT
        user_id,
        category,
        DENSE_RANK() OVER(PARTITION BY user_id ORDER BY save_count DESC) as rnk
    FROM user_category_counts
)
SELECT * FROM ranked_categories WHERE rnk <= 3;

Key Checkpoints:

• Did you filter by action_type = 'save'?
• Did you use DENSE_RANK vs RANK? (Handle ties correctly).
• Did you optimize the JOIN? (Filter timestamp before join if possible).

Module 3: Experimentation (A/B Testing)

由于 Pinterest 的 Feature 往往互相影响 (Network Effect)，A/B Testing 很难。

• Challenge: 如果你改变了 recommendation algorithm，Control Group 的用户也可能因为 Treatment Group 用户的行为改变而受到影响 (Spillover Effect)。
• Solution: 使用 Cluster Randomization (switchback testing) 或者 User-side Experimentation with careful isolation.

[!TIP] Expert Insight: 在回答 A/B Testing 问题时，不要只背诵 "Sample Size" 公式。要展现你对 "Interference" 和 "Novelty Effect" 的理解，这才是 Senior 和 Junior 的分水岭。

总结

准备 Pinterest 面试，请多刷 SQL (Road to 100 题)，并深入阅读 Pinterest Engineering Blog。懂技术，更要懂 "Inspiration"。