Towards Cheaper Tourists' Emotion and Satisfaction Estimation with PCA and Subgroup Analysis

Smart tourism leverages ubiquitous sensors to recognise the state of tourists and provide them with a better-tailored sightseeing experience. We previously reported on our EmoTour system [1], which uses behavioural cues and audiovisual data collected during sightseeing to estimate tourists' emotional status and satisfaction levels. Some of this data is however not exceedingly convenient to collect, as eye-gaze trackers for instance are not widely available nor usually worn by regular tourists. In this paper, we explore different possibilities to both improve our previous results and lessen the cost of data collection, to work towards a system that is better suited for real-world applications. Using Principal Component Analysis dimensionality reduction, we show how leaving out either or both of eye-gaze tracker and physiological wristband sensor data can have little to no impact on the quality of predictions, and improve on our previously reported classification and regression scores. We also apply this new method to explore differences in emotional responses according to participants' nationality, age, and gender.