PracHub
QuestionsLearningGuidesInterview Prep
|Home/Machine Learning/Meta

Determine Features for Effective Hashtag Recommendations

Last updated: Mar 29, 2026

Quick Overview

Evaluates hashtag recommendation system design across features, cold start, scoring, and learning. Strong answers separate candidate generation from ranking and include content, user, graph, trend, safety, and evaluation signals.

  • medium
  • Meta
  • Machine Learning
  • Data Scientist

Determine Features for Effective Hashtag Recommendations

Company: Meta

Role: Data Scientist

Category: Machine Learning

Difficulty: medium

Interview Round: Onsite

##### Scenario Designing a hashtag recommendation system for a social-media platform ##### Question What signals or features would you collect to recommend hashtags to users? For users where those features are unavailable or uninformative (e.g., new users), how would you handle recommendations? How would you combine the collected features into a scoring function? How would you determine or learn the weights for each feature? ##### Hints Consider engagement history, content similarity, demographics, popularity trends, and weight learning via regression or online learning.

Quick Answer: Evaluates hashtag recommendation system design across features, cold start, scoring, and learning. Strong answers separate candidate generation from ranking and include content, user, graph, trend, safety, and evaluation signals.

Related Interview Questions

  • Self-Attention: Implementation, Complexity, and Efficient Variants - Meta (hard)
  • Machine Learning Fundamentals: Optimizers, Scaling Laws, and Clustering - Meta (hard)
  • Implement 1NN Embeddings and Forward Pass - Meta (hard)
  • Design and evaluate an ads ranking algorithm - Meta (easy)
  • How would you design a Shop Ads ranking algorithm? - Meta (easy)
|Home/Machine Learning/Meta

Determine Features for Effective Hashtag Recommendations

Meta logo
Meta
Jul 12, 2025, 6:59 PM
mediumData ScientistOnsiteMachine Learning
105
0

Hashtag Recommendation System Design

You are designing a hashtag recommendation system for a social-media platform. Given a user composing post content at a specific time, the system should rank and recommend the top-k hashtags.

Constraints & Assumptions

  • Separate candidate generation from ranking.
  • Include user, content, social, temporal, popularity, and safety signals.
  • Address cold start for new users, new posts, and new hashtags.
  • Define how model weights or parameters are learned and validated.

Clarifying Questions to Ask Guidance

  • What is the primary objective: hashtag adoption, post engagement, discovery, creator satisfaction, or safety?
  • What content modalities are available: text, image, video, audio, or metadata?
  • Are recommendations shown during composition, after posting, or both?
  • What safety, spam, abuse, or policy constraints apply to hashtags?

Part 1 - Signals and Features

What signals would you collect to recommend hashtags for a given user, content, and time?

What This Part Should Cover Guidance

  • Include content similarity, text embeddings, image/video labels, user history, social graph, location, language, time, trending topics, hashtag popularity, engagement, and safety signals.
  • Distinguish candidate-generation features from ranking features.
  • Include freshness and decay for trends.
  • Avoid leakage from future engagement.

Part 2 - Cold Start

How would you handle users or hashtags with unavailable or weak features?

What This Part Should Cover Guidance

  • Use content-based retrieval, global or local trends, onboarding interests, similar users, and contextual defaults.
  • For new hashtags, use text semantics, early engagement, creator quality, and safety checks.
  • Explore new items with guardrails.
  • Fall back gracefully when personalization is weak.

Part 3 - Scoring and Learning

How would you combine features into a ranked list and learn weights?

What This Part Should Cover Guidance

  • Use a scoring function, learning-to-rank model, logistic model, gradient-boosted trees, neural ranker, or hybrid approach.
  • Train on outcomes such as acceptance, post engagement, downstream discovery, or long-term value.
  • Include negative sampling, calibration, diversity, deduplication, and business rules.
  • Validate offline and online with appropriate metrics.

Follow-up Questions Guidance

  • How would you prevent recommending spammy or harmful hashtags?
  • How would you measure recommendation quality if users ignore all suggestions?
  • What if a trending hashtag is popular but irrelevant to the user's post?
Loading comments...

Browse More Questions

More Machine Learning•More Meta•More Data Scientist•Meta Data Scientist•Meta Machine Learning•Data Scientist Machine Learning

Write your answer

Your first approved answer each day earns 20 XP.

Sign in to write your answer.
PracHub

Master your tech interviews with 9,000+ real questions from top companies.

Product

  • Questions
  • Learning Tracks
  • Interview Guides
  • Resources
  • Premium
  • For Universities

Browse

  • By Company
  • By Role
  • By Category
  • Topic Hubs
  • SQL Questions
  • AI Coding Questions
  • Compare Platforms
  • Discord Community

Support

  • support@prachub.com
  • (916) 541-4762

Legal

  • Privacy Policy
  • Terms of Service
  • About Us

© 2026 PracHub. All rights reserved.