Sentiment Analysis Based on Smart Human Mobility: A Comparative Study of ML Models

The great social development of the last few decades has led more and more to free time becoming an essential aspect of daily life. As such, there is the need to maximize free time trying to enjoy it as much as possible and spending it in places with positive atmospheres that result in positive sentiments. In that vein, using Machine Learning models, this project aims to create a time series prediction model capable of predicting which sentiment a given place cause on the people attending it over the next few hours. The predictions take into account the weather, whether or not an event is happening in that place, and the history of sentiment in that place over the course of the previous year. The extensive results on dataset illustrate that Long Short-Term Memory model achieves the state-of-the-art results over all models. For example, in multivariate model, the accuracy performance is 80.51% when it is applied on the LinkNYC Kiosk dataset.