Netflix is trying to predict what you will watch next. Amazon what you will buy. Facebook whom you would add next.
Recommendations are predictions, and these predictions are key for increasing user engagement.
There are two main approaches to recommendations:
- Collaborative filtering: use a combination of historical data and the Nearest Neighbors algorithm. It is based on the assumption that the user, and similar users, like what they’ve liked.
- Content-based filtering: by extracting an item’s properties (using the tf-idf algorithm) that a user liked it, the recommendation system will search for items with similar features.
What is the best one? The answer is both. In 2009, AT&T engineers won the $1M Netflix Prize, by blending multiple individual predictors into a single solution.
Founder & CEO at InTheValley
I build remote engineering teams for startups. My blog uses a proprietary platform to generate in-depth articles about something I’m passionate about: Advertising.
Latest posts by Leo Celis (see all)
- Human-in-the-Loop for Bias Mitigation - 10/16/24
- Challenges in Implementing Federated Learning in Ad Tech - 10/09/24
- Graph Neural Networks for RTB Systems - 10/02/24