Recommendation Systems Explained

Article 10: Recommendation Systems – Collaborative Filtering, Factorization

Recommendation systems shape user experiences across streaming services, online stores, and social media platforms. The core task is to balance relevance, novelty, and diversity while managing performance constraints.

Collaborative filtering utilizes techniques like user-based or item-based approaches, cosine similarity, and kNN algorithms. Matrix factorization methods include SVD, implicit feedback (ALS/BPR), and regularization strategies. Hybrid systems combine content features with collaborative signals.

Features: History, Context, Temporal Signals, Positional Effects

Cold start problems arise when dealing with new users or items; solutions often involve leveraging content-based features. Exposure bias can amplify popular elements, while signal contamination from bots and manipulation poses a challenge.

Strong recommendations rely on an architecture of ‘candidates plus ranking,’ combined with accurate metrics and ethical user interaction practices rather than solely high model accuracy.

live demo · related simulation● LIVE

Recommendation Systems Must Solve Multi-Criteria Optimization

1) Collect signals: gather data on views, clicks, purchases, and dwell time. 2) Create a candidate generator (ALS/ANN). 3) Build a ranking model using list metrics (NDCG). 4) Implement online experiments and monitor filter bubbles.

A two-tower model employs separate embeddings for users and items, utilizing cosine similarity for candidate generation followed by a neural ranker with contextual features (time, platform) for personalization.

Frequently asked questions

How can exploration strategies like epsilon-greedy be implemented in recommendation systems?

Exploration techniques, such as epsilon-greedy or Thompson sampling, can be integrated to introduce randomness and prevent models from getting stuck recommending only popular items. This helps discover potentially relevant items that the user might not have considered otherwise.

What methods are used to control diversity and novelty in recommendation systems?

Diversity and novelty can be controlled through techniques like Thompson sampling, bandit algorithms, or by explicitly penalizing items frequently recommended. These strategies encourage the system to present a wider range of options.

How are long-term metrics like retention evaluated in recommendation systems?

Long-term metrics, such as user retention (the percentage of users who continue using the system over time), are typically assessed through cohort analysis and tracking user engagement patterns. Monitoring these metrics provides insight into the overall effectiveness of the recommendations.

What strategies can be employed to filter out spam, bots, and manipulative behaviors in recommendation systems?

Filtering techniques include anomaly detection algorithms, content analysis, and user feedback mechanisms. These measures help mitigate the negative impacts of malicious actors and ensure a trustworthy recommendation experience.

▶ Try it live

Everything above runs in your browser — open Hash Function Avalanche Visualizer and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.

▶ Open Hash Function Avalanche Visualizer simulation