How recommendation engines choose what to show users, from collaborative filtering to modern ranking models.
57 terms
Session-based recommendation predicts the next item a user wants from the clicks in the current session alone, without a long-term user profile.
A shilling attack injects fake user profiles into a recommender system to push target items up or down in the recommendations shown to real users.
Top-N recommendation is the recommender systems task of returning a ranked list of the N items each user is most likely to engage with, judged by…
A two-tower model encodes users and items in separate neural networks so recommender systems can retrieve relevant candidates from millions of items quickly.
User-based collaborative filtering recommends items by finding users with similar rating histories and suggesting what those nearest neighbors liked.
A user-item matrix records how each user interacted with each item, forming the core data structure behind collaborative filtering in recommender systems.
Wide & Deep is a ranking model for recommender systems that jointly trains a linear model on crossed features and a deep neural network on embeddings.