Resumen
Recommender systems were one of the first mass applications of data analysis in various fields. The reason is their final result (recommendations) that is transparent to end users and clear metrics for measuring the quality of their work. End-users can always evaluate the usefulness of recommendations, formal measurements can always operate on conversion, whatever it means - purchases of recommended products, clicks on links, etc. Most often, the work of recommender systems is based on the generalization and analysis of the preferences of other users (which includes consideration of various aspects of their behavior), and the available information about the current user. At the same time, there is a class of tasks when recommendations should (or only can) be based on the current actions of the user. For example, in an e-commerce system, an unauthorized (anonymous) user visits various pages of a site. Or the user's preferences in the system are only short-term. All these examples are typical for a separate large class of recommender systems - recommender systems for sessions, where a session is understood as a sequence of user actions. The recommender system in this case solves one of three tasks: recommends the next product (content, activity, etc.) within the current session, recommends the following products (activities, etc.) until the end of the current session, recommends the next possible session. The article contains an overview of the described tasks and models for such recommender systems.