Cyclist Experience Sampling in the Wild: A Memory-aware Sentiment Strength Extraction Method

Experience sampling methods aim to obtain self-report and momentary cues of emotion in context. However, in the context of driving, it might be dangerous to ask a driver to answer a question in the moment. Therefore, we propose an image sentiment strength calculation method that considers the effect...

Full description

Saved in:
Bibliographic Details
Published in2022 10th International Conference on Affective Computing and Intelligent Interaction Workshops and Demos (ACIIW) pp. 1 - 2
Main Authors Dollack, Felix, Li, Da, Yamaguchi, Ryuta, Yoshihisa, Tomoki, Shimojo, Shinji, Kawai, Yukiko
Format Conference Proceeding
LanguageEnglish
Japanese
Published IEEE 18.10.2022
Subjects
Affective computing
Conferences
driving
Faces
feedback
image processing
memory
Real-time systems
Roads
Sampling methods
sentiment strength
Vehicles
Online AccessGet full text

Cover

More Information
Summary:Experience sampling methods aim to obtain self-report and momentary cues of emotion in context. However, in the context of driving, it might be dangerous to ask a driver to answer a question in the moment. Therefore, we propose an image sentiment strength calculation method that considers the effect of memory. We applied a real-time face analysis API to obtain facial expression labels from images taken during driving. Then, we extracted the road images using the GPS coordinates where the facial expression images were taken. Afterwards, we presented the road images to the users and asked them to report their sentiment at the moment they passed the location. This self-report served as ground truth sentiment. Furthermore, we considered the impact of memory on the self-report. We sent each extracted road image to users for evaluation after certain time had elapsed after riding to acquire a self-report. We then calculated a weighted sum to represent the impact of memory according to the self-report scores received at different times. Finally, we combined the weighted self-report and face analysis API scores to calculate the final sentiment report.
DOI:10.1109/ACIIW57231.2022.10086030