POI-RGNN: Using Recurrent and Graph Neural Networks to Predict the Category of the Next Point of Interest

Abstract

Recommendation systems have been used to predict the next place a user is willing to visit. However, these methods commonly achieve low hit rates because they aim to recommend exact locations among many possibilities. A higher-level approach that is more effective is to predict the category of the next location since it can be helpful in a variety of services. For example, it is possible to do category-based location recommendations, more assertive advertising programs, among others. In this work, we present POI-RGNN (Points of Interest (PoI) - Recurrent and Graph-based Neural Network), a neural network for predicting the category of the next PoI that an individual will visit. Our proposal leverages Recurrent Neural Networks (RNN) and Graph Neural Networks (GNN) and combines them in a novel architecture. Additionally, the POI-RGNN explores new types of inputs that are sent to recurrent and graph layers. Results show that the proposed model improves macro and weighted f1-score among all PoI categories. We evaluate POI-RGNN in two distinct types of real-world datasets, showing its effectiveness in different contexts.