Analyze View Distribution and Recommendation Overlap in Videos
Overview
This question evaluates statistical competence in analyzing heavy-tailed view distributions and measuring recommendation overlap, testing skills such as data visualization, computation and interpretation of mode/median/mean/99th percentile, and statistical assessment of overlap in top-10 video lists for a Data Scientist.
Meta
Aug 4, 2025, 10:55 AM
Data Scientist
Onsite
Statistics & Math
0
0
Short-Video Platform: View Distribution and Recommendation Overlap
Context
You are analyzing a short-video platform. You have:
- A dataset of per-video view counts over a fixed time window (e.g., last 30 days).
- Two users whose top-10 recommended videos (or top-10 consumed videos) frequently include identical items.
Assume view counts are nonnegative integers and video identity is deduplicated (e.g., by content hash, not just URL) to avoid counting re-uploads separately.
Tasks
- Distribution of video-level views
- Describe how you would visualize the distribution of views per video.
- Report the mode, median, mean, and 99th percentile of the distribution.
- Overlap in top-10 videos between two users
- Statistically evaluate whether frequent overlap in two users' top-10 lists is desirable or a potential problem.
- Explicitly consider heavy-tail effects (Zipf-like distributions), and discuss trade-offs between diversity and homogeneity.
Hints
- Expect a heavy-tailed, long-tail distribution (often Zipf/Pareto-like).
- Weigh personalization and diversity against the benefits of showing trending, high-quality content.
Solution
Login required
Show