Question Bank
RecSys interview questions
Recommendations must find relevant items while respecting product constraints. Review candidate retrieval, ranking and evaluation. Explain how cold starts, user feedback and the difference between offline metrics and online experiments affect your choices.
Showing 20 of 139 questions
Explain the concept of cold start in recommendation systems and how it can be mitigated.
Why can the position of recommendations affect user engagement?
What are the key principles of using graphs in recommendation systems?
Compare nearest neighbor methods such as FAISS and Annoy in terms of their performance and application.
Describe a recommendation project in detail: problem definition, testing, and challenges.
How did you determine which product categories are compatible for recommendations?
What features did you use in the recommendation ranking model?
How did you build product embeddings for the recommendation system?
Explain ranking metrics: Precision@K, Recall@K, NDCG, and MRR.
Which business metric would you optimize in a short-video recommendation system: average session time or total session time?
Why not simply rank by the probability output of a binary classifier? What advantage does pairwise ranking offer?
Why can't NDCG be optimized directly as a loss function? What alternatives are available?
Explain singular value decomposition (SVD) and how it is used in recommendation systems.
How would you use content-based and collaborative embeddings for candidate generation in a recommendation system?
How would you address the cold-start problem in recommendation systems?
Which ranking metrics do you know, and how do they differ?
How would you improve a recommendation algorithm as more data becomes available?
Compare pointwise and pairwise ranking approaches in terms of quality, speed, and metrics.
Prepare for your next interview with Vibe Interview.
Download Vibe Interview