Emotion predictions in geo-textual data using spatial statistics and recommendation systems

Abstract

Microblogs are used by millions of users to express their emotions, such as joy, surprise and anger, on a plethora of different topics. For the same topic, different places may exhibit different emotions for identical topics. The goal of this work is to learn, model and predict emotions on various topics and in different cities. For this purpose, we propose a hybrid approach which combines spatial statistics (kriging) and recommendation system (matrix factorization-based). Our experimental evaluations, using millions of tweets across the United States, show that our hybrid approach outperforms individual approaches based on matrix factorization and Kriging alone. This case study shows the potential of combining spatial statistics methods such as Kriging with machine learning solutions to support knowledge discovery on spatial data.