Wang et al. present a comprehensive survey with this paper. A session-based recommender system (SBRS) is a system that makes recommendations to users based on short-term, dynamic user preferences (in a session). It is different from other recommendation systems, such as content-based and collaborative filtering-based recommender systems, which usually model long-term, static user preferences.

The key components of an SBRS include users, items, and actions. An SBRS is a collection of interactions consisting of triplets u, v, a: a user u takes an action a on an item v. A session is a non-empty, time-bounded list of interactions. The way an SBRS solves a problem is to maximize the utility score conditioned on the given session context. For the given session information, an SBRS attempts to produce a predicted list of interactions.

An SBRS can be classified into three groups based on the approaches used: 1) conventional approaches, including pattern/rule mining, k-nearest neighbor, Markov chains, and generative probabilistic models; 2) latent representation-based approaches, including latent factor models and distributed representation; and 3) deep neural network (DNN)-based approaches. An SBRS can be applied to conventional applications such as E-commerce, media, entertainment, and tourism. They can also be applied to emerging areas such as finance and healthcare.

According to the authors, the research on session-based recommender systems is flourishing and new techniques and approaches continue to be developed.

This is a very comprehensive survey paper with 153 listed references. Readers can benefit from the completeness and easy-to-read nature of the paper. It is especially appropriate for readers who would like to learn more about the general concepts in the area.