Desarrollo de un modelo adaptativo-dinámico de recomendación para grupos basado en aspectos temporales.

Abstract

Recommender systems (RS) implemented on Web platforms have reached a certain maturity and today, they are capable of finding the products or services that best suit the needs and interests of potential users or groups of users. Actually, group recommender systems (GRS) allow items to be recommended to a group of users who have common interests. The temporal aspects of GRS are based on the idea of generating a sequence of ratings where each rating belongs to a temporary space or period, so group interest drift is captured over time. Therefore, a temporal-sensitive GRS seeks to predict the group-item rating for the next time period in the future. As in other contexts, this also allows the generation of recommendation lists for groups. The approach on sequential aspects used for the design of a SRG considering the temporal aspects, is the one proposed in this research. This is based on the evaluation of some methods. Indeed, we used matrix factorization for individual rating predictions to later detect groups and extract the group models with strategies of aggregation. Finally, state-of-the-art sequence-aware RS model were applied in a GRS scenario. MovieLens and Amazon datasets having the timestamp field were selected to be able to evaluate the models. Metrics such as the RMSE and NDCG were reported.