1139 | 837 | The potential implications of the urban environment on human sentiment

Mental health problems have been on the rise worldwide, possibly associated with urban population growth and lifestyles. The recognition that the various aspects of the urban environment can affect the mental health of individuals has been increasing since they are responsible for facilitating or inhibiting behaviors and lifestyles that impact the feeling._x000D_
In this context, it is essential to understand the potential impact that the urban environment of the city of Lisbon may have on the feelings of those who “live the space”. To do so, we resorted to sentiment analysis based on data from the social network Twitter, using a lexicon from the NRC Sentiment and Emotion, enabling the identification of places where both positive and negative sentiment prevail; this is an easily replicable process with a more direct association to sentiment and emotions (e.g., Plutchik’s wheel of emotions). _x000D_
Next, a Machine Learning (ML) model associated with an agnostic model was used to increase understanding of the factors in the urban environment that can explain sentiment (hedonic well-being). Four ML models were tested, Random Forest (RF), Extreme Gradient Boosting (XGBoost), Neural Network (NN), and K-Nearest Neighbour (KNN), which is one of the simplest algorithms used, and a linear model for comparison (Generalized Linear Model – GLM). Using positive/negative sentiment as dependent variable and 30 explanatory variables related to the urban environment, it was found that RF is the model with the highest predictive ability._x000D_
The agnostic models applied, the Local Interpretable Model-Agnostic Explanations (LIME), and the SHapley Additive exPlanation (SHAP), which is based on game theory, played a crucial role in this study. Thus, answering the starting question, the explanatory variables that are most related to sentiment are distance to fitness equipment, distance to green spaces, the popularity of locations, and distance to the cycling network._x000D_

Iuria Betco
Centro de Estudos Geográficos, Instituto de Geografia e Ordenamento do Território, Universidade de Lisboa, Portugal

ID Abstract: 837