Exploring temporal contexts for neighborhood-based models in Session-Based Recommender Systems
Gabriel B. Domingos, Igor Lovatto, Matheus Naoto Shimura Takata, Marcelo Garcia Manzato · 2019
Recent works in the field of Recommender Systems have focused in making recommendations using only information about accessed items in an active user session, since the user’s preferences can be very specific to the context of the session. Despite the increasing interest in deep neural network models in the Recommender Systems in general, simpler models, such as neighborhood-based collaborative filtering ones, have been outperforming these more complex models in the session-based scenario. However, in order to soften known scalability problems in neighborhood-based models, sampling strategies during the recommendation process have been proposed, showing that some session contexts such as recency of neighbor sessions can improve the recommendation in some domains. Therefore this project explores the effect of different temporal contexts that can easily be captured from an anonymous user session, such as the hour of the day and day of the week, as alternatives to the established sampling methods.