Skip to main content

Recommender System

A Recommender System (a.k.a. Collaborative Filtering, Social Filtering, and Social Information Filtering) is an information filtering technique that takes details and data associated with a user's profile and compares it with similar data (habits, likes, opinions, etc.) of other users on the same service in order to present recommendations of what might be of interest to the original user.

Examples of web services that make use of Recommender Systems are: online auction/ecommerce websites, music services, movie and television show streaming services, etc.

  • Frequently bought together for Retail
  • Top picks for you for Media and Entertainment

https://www.quora.com/LinkedIn-Recommendations/How-does-LinkedIns-recommendation-system-work

https://docs.aws.amazon.com/personalize/latest/dg/what-is-personalize.html

Vinija's Notes • Recommendation Systems • Research Papers

Social Media Recommendation Engine​

  • Doom scrolling
  • Endless bottom / endless scrolling
  • Reel life vs real life

Two-Tower Model for Recommendation Systems​

The two-tower approach is a deep learning architecture in recommendation systems that uses two separate neural networks, or "towers," to generate embeddings (vector representations) for users and items independently. During training, the towers learn to produce embeddings such that user and item embeddings for positive interactions are close in a shared latent space, allowing for efficient similarity calculation (e.g., dot product) to predict relevance. This decoupling of user and item processing enables the pre-computation of item embeddings for fast, large-scale candidate retrieval in real-time.

The Two-Tower Model for Recommendation Systems: A Deep Dive | Shaped Blog

Price Recommendation Engine​

ML driven dynamic pricing @ CARS24 - Part 1 | by Shashank Kumar | CARS24 Data Science Blog | Mar, 2023 | Medium