DataTerminology.comNEW
Also known as: Recommender
Software that predicts user preferences and suggests relevant items, optimizing engagement and utility.
DESCRIPTION
Recommendation systems filter information to present items users are likely to value. Core approaches include collaborative filtering, content-based methods, and hybrid models. Signals can be explicit ratings or implicit behaviors like clicks and dwell time. Modern recommenders use embeddings, sequence models, and contextual bandits. They address cold start via metadata and side information. Objectives balance relevance, diversity, novelty, and fairness. Evaluation combines offline metrics with online A/B testing. Production setups involve candidate generation, scoring, and ranking stages.
ALTERNATIVE DEFINITIONS
Wikipedia
Recommender systems are a subclass of information filtering systems that seek to predict the rating or preference a user would give to an item.
Emphasizes prediction of user-item preferences within information filtering.
Google Developers
Recommender systems predict what users will like based on past behavior and similar users.
Highlights collaborative and behavioral signals for recommendation.
Microsoft Learn
A recommendation engine suggests relevant items to users by analyzing usage patterns and item attributes.
Positions recommenders as a practical solution using behavior and metadata.
RELATED TERMS
Collaborative Filtering Content-Based Filtering Matrix Factorization Implicit Feedback Initialization Personalization Ranking Exploration-Exploitation Contextual Bandits User Segmentation
COMPETING TERMS
Search Engine Manual Curation
KNOWLEDGE AREAS
TARGET AUDIENCE
AI Engineer Data Scientist Data Engineer AI Product Manager
EXTERNAL SOURCES
- Wikipedia - Recommender system
- Google Developers - Overview of recommender systems
- AWS - Amazon Personalize product overview
- Netflix Research - Personalization research area
TOP QUESTIONS
What is a Recommendation System?
It predicts user preferences and suggests items likely to satisfy them, using signals like past behavior, item attributes, and similarities among users and items.
Category: Definition
What are common algorithms in Recommendation Systems?
Collaborative filtering, content-based filtering, matrix factorization, sequence models, graph-based recommenders, and contextual bandits are commonly used.
Category: Concepts
Why are Recommendation Systems important?
They boost engagement, revenue, and user satisfaction by surfacing relevant items, enabling discovery, and reducing information overload in large catalogs.
Category: Relevancy
How do Recommendation Systems handle cold start?
They leverage metadata, item content, demographic data, and popularity priors. Active learning and onboarding questions also help mitigate cold start.
Category: Requirements
What metrics evaluate Recommendation Systems?
Offline metrics include precision@k, recall@k, NDCG, coverage, and diversity. Online A/B tests track CTR, conversion, dwell time, and long-term engagement.
Category: Testing
What risks exist with Recommendation Systems?
Filter bubbles, popularity bias, and unfair treatment of items or creators can arise. Poor guardrails may promote low-quality or harmful content.
Category: Risks
How is a Recommendation System architected?
Typical stacks have candidate generation, scoring, and ranking layers, supported by feature stores, model registries, and real-time feedback pipelines.
Category: Architecture
How do Recommendation Systems balance exploration and exploitation?
Bandit algorithms and controlled diversification introduce exploration, ensuring discovery without sacrificing short-term relevance too much.
Category: Opinion
Where to start learning Recommendation Systems?
Begin with collaborative filtering and matrix factorization, then study embeddings, sequence models, and online testing. Vendor docs and industry blogs offer practical guidance.
Category: GettingStarted