juniorRecommendation Systems

What are the different types of recommendation approaches, such as collaborative filtering and content-based filtering?

Updated Feb 20, 2026

Short answer

Recommendation systems suggest items to users by learning preferences from user behavior, item characteristics, or both. The main approaches are collaborative filtering, content-based filtering, hybrid methods, and knowledge-based or context-aware recommendations. Collaborative filtering uses patterns across users, while content-based filtering uses similarities between items and a user's past interests.

Deep explanation

A recommendation system is a machine learning system that predicts which items a user is likely to interact with, purchase, watch, or enjoy. The goal is to rank possible items and present the most relevant ones.

The major recommendation approaches are:

1. Collaborative Filtering

Collaborative filtering recommends items based on the behavior of many users. The key idea is:

Users who behaved similarly in the past may like similar things in the future.

It does not require understanding the actual content of items. Instead, it learns from interactions such as:

  • Ratings
  • Clicks
  • Purchases
  • Watch history
  • Likes
  • Search behavior

There are two main types:

User-based Collaborative Filtering

This approach finds users who are similar to the target user and recommends items those similar users liked.

Example:

  • Alice and Bob both watched many of the same movies.
  • Bob watched a new movie that Alice has not seen.
  • The system recommends that movie to Alice.

A simplified process:

  1. Create a user-item interaction matrix.
TypeScript
Movie A   Movie B   Movie C
Alice           5         4         ?
Bob             5         4         5
Carol           1         2         1
  1. Calculate similarity between users.
  2. Find users with similar preferences.
  3. Recommend items liked by those users.

Item-based Collaborative Filtering

Instead of finding similar users, this approach finds similar items.

Example:

  • Users who watched "The Matrix" also frequently watched "Inception".
  • A user who liked "The Matrix" may receive "Inception" as a recommendation.

Many large-scale systems prefer item-based methods because item relationships are often more stable than user relationships.

Advantages of Collaborative Filtering

  • Does not require manual item descriptions.
  • Can discover unexpected recommendations.
  • Learns complex user preferences automatically.

Limitations

  • Cold start problem: New users or new items have little interaction data.
  • Data sparsity: Most users interact with only a tiny fraction of available items.
  • Popularity bias: Popular items may dominate recommendations.

---

2. Content-Based Filtering

Content-based filtering recommends items based on the characteristics of items and the user's previous preferences.

The key idea is:

Recommend items similar to what the user already liked.

The system creates a profile of the user's interests from past interactions.

For example, for movies, item features may include:

TypeScript
Movie:
{
  genre: ["science fiction", "action"],
  director: "Christopher Nolan",
  language: "English",
  year: 2010
}

If a user frequently watches science fiction movies, the system recommends other science fiction movies.

A typical pipeline:

  1. Extract item features.
  2. Build a user preference profile.
  3. Calculate similarity between the user profile and candidate items.
  4. Rank the items.

Common similarity techniques include:

  • Cosine similarity
  • Jaccard similarity
  • Neural embeddings

Example:

TypeScript
User Profile:
[
  "science fiction": 0.8,
  "action": 0.6,
  "romance": 0.1
]


Candidate Movie:
[
  "science fiction": 0.9,
  "action": 0.7,
  "romance": 0.0
]


Similarity score = high

Advantages of Content-Based Filtering

  • Works well for new users if they provide preferences.
  • Does not require data from other users.
  • Recommendations are easier to explain.

Limitations

  • Can become too narrow.
  • Depends on quality of item features.
  • May fail to discover unexpected interests.

---

3. Hybrid Recommendation Systems

Modern recommendation systems usually combine multiple approaches.

A hybrid system may combine:

  • Collaborative filtering signals
  • Content features
  • User demographics
  • Context information
  • Deep learning models

Example:

A streaming platform may consider:

TypeScript
Recommendation Score =
0.5 * User Similarity Score
+ 0.3 * Content Similarity Score
+ 0.2 * Trending Score

Benefits:

  • Reduces cold-start problems.
  • Improves recommendation accuracy.
  • Balances personalization and popularity.

Many large platforms use hybrid architectures because no single method works well in every situation.

---

4. Knowledge-Based Recommendation

Knowledge-based systems recommend items using explicit rules and domain knowledge.

They are useful when:

  • Purchases are rare.
  • Items are expensive.
  • User preferences are highly specific.

Examples:

  • Real estate recommendations.
  • Car recommendations.
  • Financial product recommendations.

A user might specify:

TypeScript
Budget: $50,000
Location: New York
Bedrooms: 3+

The system filters and ranks properties matching those requirements.

---

5. Context-Aware Recommendation

Context-aware systems include information beyond users and items.

Context can include:

  • Time
  • Location
  • Device
  • Weather
  • Current activity

Example:

A food delivery app may recommend:

  • Coffee shops in the morning.
  • Dinner restaurants in the evening.
  • Nearby restaurants during lunchtime.

---

Comparison of Approaches

ApproachUsesStrengthWeakness
Collaborative FilteringUser interactionsDiscovers hidden patternsCold start problem
Content-Based FilteringItem features + user historyPersonalized and explainableLimited discovery
HybridMultiple signalsUsually best performanceMore complexity
Knowledge-BasedRules and constraintsWorks without much historyRequires domain knowledge
Context-AwareSituation informationMore relevant recommendationsRequires additional data

In practice, recommendation systems often have several stages:

  1. Candidate generation: Quickly retrieve hundreds of possible items.
  2. Ranking: Use machine learning models to score candidates.
  3. Filtering: Apply business rules and diversity constraints.

A simplified architecture:

TypeScript
User Request
      |
      v
Candidate Generation
      |
      +--> Collaborative Filtering
      |
      +--> Content Matching
      |
      v
Ranking Model
      |
      v
Final Recommendations

Real-world example

Consider a video streaming service recommending movies.

A hybrid recommendation system might work like this:

  1. Collaborative filtering finds movies watched by users with similar viewing history.
  2. Content-based filtering finds movies similar to ones the user already watched.
  3. A ranking model combines the results.

Example:

Python
def recommend_movies(user):
    collaborative = get_similar_user_movies(user)
    content_based = get_similar_movies(user.favorite_movies)


    candidates = collaborative + content_based


    ranked = rank_movies(
        candidates,
        user_preferences=user.preferences
    )


    return ranked[:10]

If a user watches many science fiction movies, the system may recommend:

  • Similar science fiction movies from content matching.
  • Movies enjoyed by users with similar viewing patterns from collaborative filtering.
  • Popular movies currently trending.

The final list is usually produced by combining all these signals.

Common mistakes

  • * Assuming collaborative filtering understands the meaning of items
  • it only learns patterns from interactions.
  • * Assuming content-based filtering can discover completely new interests without additional signals.
  • * Ignoring the cold-start problem for new users and new items.
  • * Using only popularity ranking and calling it personalization.
  • * Measuring success only with accuracy instead of considering diversity, novelty, and user satisfaction.
  • * Forgetting that recommendation systems can reinforce biases by repeatedly showing similar content.
  • * Building a complex deep learning model before establishing strong baseline recommendation approaches.

Follow-up questions

  • How does collaborative filtering handle the cold-start problem?
  • What is the difference between user-based and item-based collaborative filtering?
  • Why are hybrid recommendation systems commonly used in production?
  • How do recommendation systems evaluate their performance?
  • What is the role of machine learning models in modern recommendation systems?

More Recommendation Systems interview questions

View all →