Sentiment Analysis Models with Bayesian Approach: A Bike Preference Application in Metropolitan Cities

Abstract

Social media data is an important source of information that can also be used for the study of the passenger mobility sector. In transport systems, user choice is studied through demand models that define how user behavior is affected by the performance of the supply system. Demand models are typically calibrated through data observed in the transport system. The observed data includes the choices actually made by users. This paper investigates how sentiment analysis of data available in social media can be adopted to specify, calibrate, and validate demand models in certain choice levels. In this work a model based on the Bayesian approach is specified, calibrated, and validated in the case of bike preference in some Italian metropolitan cities. The model takes into account the discrete choice approach. Specification, calibration, and validation made it possible to identify the relevant variables that influence sentiments and obtain the posterior distribution probability of the parameters. The prior and the posterior conditional probabilities are compared, and some indications are obtained on the elasticity and weight of the sentiments that influence the choice.