Recommender system

A software system that predicts which items, services, or information a user may prefer and presents personalized suggestions. It uses data about users, items, or interactions to rank or select recommendations.

Collaborative filtering: A recommendation method that predicts preferences from patterns in user-item interactions. It recommends items by finding useful similarities among users or their interaction histories.
E-commerce product recommendations: Personalized suggestions of products in online shopping services. Shopping systems use recommendations to help customers discover products and increase purchases.
Filter bubble: A pattern in which personalized information environments limit exposure to diverse viewpoints or material. Repeatedly optimizing for predicted interests can narrow the range of content shown.
Non-personalized recommendation: Suggestions selected without tailoring them to an individual user's preferences. Global bestsellers or trending lists provide a baseline without user-specific prediction.
Differential privacy: A mathematical framework for limiting what an analysis reveals about any individual in its dataset. It offers formal privacy protections for learning from user interaction data.
Content-based filtering: A recommendation method that matches item attributes to a user's recorded preferences. It can personalize suggestions from item features even when other users' behavior is unavailable.
Music recommendation: The selection of music for listeners based on preferences, context, or similarity. Music services use listening histories and audio features to assemble personalized discovery.
Algorithmic bias: Systematic and unfair outcomes produced or amplified by algorithmic systems. Training data and ranking choices can distribute recommendations unevenly across groups.
Popularity-based recommendation: A method that recommends items according to their overall popularity or activity. It is simple and useful for new users, but gives everyone largely the same suggestions.
Federated learning: A machine-learning approach that trains models across devices or organizations without centralizing raw data. It can reduce the need to collect personal histories in one central system.

Linked from 29 pages