# Recommender System

A recommender system selects and ranks items to help a user discover options likely to be useful in a particular context.

- Type: framework
- Domain: Engineering & architecture
- Level: Intermediate
- Last reviewed: 2026-10-06
- Original: [Soroush’s Product Wiki](<https://wiki.soroush.uk/topic/recommender-system/>)

## What it means

A recommender system, also called a recommendation engine, predicts which items to present using preferences, item characteristics, interaction history, or contextual signals. Content-based approaches match item attributes to interests; collaborative filtering uses patterns across users and items; hybrids combine signals. Recommendations can support discovery without an explicit query, whereas search usually begins with one. Neither personalization nor machine learning is mandatory: a curated or popularity-based list can be a useful baseline.

## Why it matters

Recommendations shape what people see, what suppliers gain exposure, and which interactions become future training data. Product decisions include the objective, eligibility rules, user controls, and the balance between familiar choices and useful discovery.

## When to use it

Use recommendations when users face many options and useful relevance signals exist. Compare against a simple baseline before adding modeling complexity. For a small catalog or an exact user request, clear navigation, filters, or search may serve the task better.

## How it works

1. Define the discovery task and desired outcome, including constraints on eligibility, repetition, and item exposure.
2. Generate a manageable candidate set using content, interaction patterns, curated rules, or a combination.
3. Score candidates for the task and re-rank to meet constraints such as availability, freshness, and diversity.
4. Provide fallbacks for new users and items with little interaction history, and let users correct unsuitable suggestions.
5. Evaluate relevance offline and customer outcomes online, checking coverage, subgroup performance, latency, and changes over time.

## A practical example

A learning platform recommends a next course from completed lessons, stated goals, and course prerequisites. New users receive goal-based suggestions because their interaction history is sparse. The team compares recommendations with a curated baseline and checks completion and satisfaction, rather than choosing the version with the most course-card clicks.

## Where it breaks

Historical interactions reflect previous exposure and may confuse popularity with preference. New-user and new-item cold starts concern missing recommendation evidence, distinct from a marketplace lacking enough participants. Offline ranking improvements may not improve customer outcomes, and personalization can narrow discovery or carry stale preferences forward.

## Common mistakes

- Optimizing clicks without checking whether recommendations help users accomplish their goals.
- Treating items a user never saw as items they disliked.
- Evaluating only established users while ignoring new users and new inventory.
- Assuming a complex personalized model must outperform a simple curated baseline.

## Visualization

[View the visualization on the website](<https://wiki.soroush.uk/topic/recommender-system/#visualization-recommender-system>)

## Sources

1. [Recommendation systems overview](<https://developers.google.com/machine-learning/recommendation/overview/types>) — Google for Developers · Official educational documentation
2. [Deep Neural Networks for YouTube Recommendations](<https://research.google/pubs/deep-neural-networks-for-youtube-recommendations/>) — ACM Conference on Recommender Systems · Peer-reviewed research
3. [Collaborative filtering advantages & disadvantages](<https://developers.google.com/machine-learning/recommendation/collaborative/summary>) — Google for Developers · Official educational documentation

## Related topics

- [Machine Learning \(ML\)](<https://wiki.soroush.uk/topic/machine-learning/>)
- [Semantic Search](<https://wiki.soroush.uk/topic/semantic-search/>)
- [Algorithmic Feedback Loop](<https://wiki.soroush.uk/topic/algorithmic-feedback-loop/>)
